# Directory Structure
```
├── .cursorrules
├── .gitignore
├── docs
│ └── mcp_spec
│ └── llms-full.txt
├── index.ts
├── LICENSE
├── package.json
├── pnpm-lock.yaml
├── README.md
└── tsconfig.json
```
# Files
--------------------------------------------------------------------------------
/.cursorrules:
--------------------------------------------------------------------------------
```
1. Use pnpm instead of npm when generating packaging-related commands.
2. Only make changes to comments, code, or dependencies that are needed to accomplish the objective defined by the user. When editing code, don't remove comments or change dependencies or make changes that are unrelated to the code changes at hand.
```
--------------------------------------------------------------------------------
/.gitignore:
--------------------------------------------------------------------------------
```
# Logs
logs
*.log
npm-debug.log*
yarn-debug.log*
yarn-error.log*
lerna-debug.log*
.pnpm-debug.log*
# Diagnostic reports (https://nodejs.org/api/report.html)
report.[0-9]*.[0-9]*.[0-9]*.[0-9]*.json
# Runtime data
pids
*.pid
*.seed
*.pid.lock
# Directory for instrumented libs generated by jscoverage/JSCover
lib-cov
# Coverage directory used by tools like istanbul
coverage
*.lcov
# nyc test coverage
.nyc_output
# Grunt intermediate storage (https://gruntjs.com/creating-plugins#storing-task-files)
.grunt
# Bower dependency directory (https://bower.io/)
bower_components
# node-waf configuration
.lock-wscript
# Compiled binary addons (https://nodejs.org/api/addons.html)
build/Release
# Dependency directories
node_modules/
jspm_packages/
# Snowpack dependency directory (https://snowpack.dev/)
web_modules/
# TypeScript cache
*.tsbuildinfo
# Optional npm cache directory
.npm
# Optional eslint cache
.eslintcache
# Optional stylelint cache
.stylelintcache
# Microbundle cache
.rpt2_cache/
.rts2_cache_cjs/
.rts2_cache_es/
.rts2_cache_umd/
# Optional REPL history
.node_repl_history
# Output of 'npm pack'
*.tgz
# Yarn Integrity file
.yarn-integrity
# dotenv environment variable files
.env
.env.development.local
.env.test.local
.env.production.local
.env.local
# parcel-bundler cache (https://parceljs.org/)
.cache
.parcel-cache
# Next.js build output
.next
out
# Nuxt.js build / generate output
.nuxt
dist
# Gatsby files
.cache/
# Comment in the public line in if your project uses Gatsby and not Next.js
# https://nextjs.org/blog/next-9-1#public-directory-support
# public
# vuepress build output
.vuepress/dist
# vuepress v2.x temp and cache directory
.temp
.cache
# Docusaurus cache and generated files
.docusaurus
# Serverless directories
.serverless/
# FuseBox cache
.fusebox/
# DynamoDB Local files
.dynamodb/
# TernJS port file
.tern-port
# Stores VSCode versions used for testing VSCode extensions
.vscode-test
# yarn v2
.yarn/cache
.yarn/unplugged
.yarn/build-state.yml
.yarn/install-state.gz
.pnp.*
```
--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
```markdown
# MCP Web Research Server
A Model Context Protocol (MCP) server for web research.
Bring real-time info into Claude and easily research any topic.
## Features
- Google search integration --- THIS FORK FIXES THIS --- NOW NO LONGER GETTING CAPTCHA BLOCKED
- Webpage content extraction
- Research session tracking (list of visited pages, search queries, etc.)
- Screenshot capture
## Prerequisites
- [Node.js](https://nodejs.org/) >= 18 (includes `npm` and `npx`)
- [Claude Desktop app](https://claude.ai/download)
## Installation
First, ensure you've downloaded and installed the [Claude Desktop app](https://claude.ai/download) and you have npm installed.
Next, add this entry to your `claude_desktop_config.json` (on Mac, found at `~/Library/Application\ Support/Claude/claude_desktop_config.json`):
```json
{
"mcpServers": {
"webresearch": {
"command": "npx",
"args": ["-y", "@mzxrai/mcp-webresearch@latest"]
}
}
}
```
This config allows Claude Desktop to automatically start the web research MCP server when needed.
## Usage
Simply start a chat with Claude and send a prompt that would benefit from web research. If you'd like a prebuilt prompt customized for deeper web research, you can use the `agentic-research` prompt that we provide through this package. Access that prompt in Claude Desktop by clicking the Paperclip icon in the chat input and then selecting `Choose an integration` → `webresearch` → `agentic-research`.
<img src="https://i.ibb.co/N6Y3C0q/Screenshot-2024-12-05-at-11-01-27-PM.png" alt="Example screenshot of web research" width="400"/>
### Tools
1. `search_google`
- Performs Google searches and extracts results
- Arguments: `{ query: string }`
2. `visit_page`
- Visits a webpage and extracts its content
- Arguments: `{ url: string, takeScreenshot?: boolean }`
3. `take_screenshot`
- Takes a screenshot of the current page
- No arguments required
### Prompts
#### `agentic-research`
A guided research prompt that helps Claude conduct thorough web research. The prompt instructs Claude to:
- Start with broad searches to understand the topic landscape
- Prioritize high-quality, authoritative sources
- Iteratively refine the research direction based on findings
- Keep you informed and let you guide the research interactively
- Always cite sources with URLs
### Resources
We expose two things as MCP resources: (1) captured webpage screenshots, and (2) the research session.
#### Screenshots
When you take a screenshot, it's saved as an MCP resource. You can access captured screenshots in Claude Desktop via the Paperclip icon.
#### Research Session
The server maintains a research session that includes:
- Search queries
- Visited pages
- Extracted content
- Screenshots
- Timestamps
### Suggestions
For the best results, if you choose not to use the `agentic-research` prompt when doing your research, it may be helpful to suggest high-quality sources for Claude to use when researching general topics. For example, you could prompt `news today from reuters or AP` instead of `news today`.
## Problems
This is very much pre-alpha code. And it is also AIGC, so expect bugs.
If you run into issues, it may be helpful to check Claude Desktop's MCP logs:
```bash
tail -n 20 -f ~/Library/Logs/Claude/mcp*.log
```
## Development
```bash
# Install dependencies
pnpm install
# Build the project
pnpm build
# Watch for changes
pnpm watch
# Run in development mode
pnpm dev
```
## Requirements
- Node.js >= 18
- Playwright (automatically installed as a dependency)
## Verified Platforms
- [x] macOS
- [x] Linux
- [x] Windows
## License
MIT
## Author
[mzxrai](https://github.com/mzxrai)
```
--------------------------------------------------------------------------------
/tsconfig.json:
--------------------------------------------------------------------------------
```json
{
"compilerOptions": {
"target": "ES2023",
"module": "NodeNext",
"moduleResolution": "NodeNext",
"esModuleInterop": true,
"strict": true,
"outDir": "dist",
"sourceMap": true,
"declaration": true,
"skipLibCheck": true,
"lib": [
"ES2023",
"DOM",
"DOM.Iterable"
]
},
"include": [
"*.ts"
],
"exclude": [
"node_modules",
"dist"
]
}
```
--------------------------------------------------------------------------------
/package.json:
--------------------------------------------------------------------------------
```json
{
"name": "@mzxrai/mcp-webresearch",
"version": "0.1.7",
"description": "MCP server for web research",
"license": "MIT",
"author": "mzxrai",
"homepage": "https://github.com/mzxrai/mcp-webresearch",
"bugs": "https://github.com/mzxrai/mcp-webresearch/issues",
"type": "module",
"bin": {
"mcp-server-webresearch": "dist/index.js"
},
"files": [
"dist"
],
"scripts": {
"build": "tsc && shx chmod +x dist/*.js",
"prepare": "pnpm run build",
"postinstall": "playwright install chromium",
"watch": "tsc --watch",
"dev": "tsx watch index.ts"
},
"publishConfig": {
"access": "public"
},
"keywords": [
"mcp",
"model-context-protocol",
"web-research",
"ai",
"web-scraping"
],
"dependencies": {
"@modelcontextprotocol/sdk": "1.0.1",
"playwright": "^1.49.0",
"turndown": "^7.1.2"
},
"devDependencies": {
"shx": "^0.3.4",
"tsx": "^4.19.2",
"typescript": "^5.6.2",
"@types/turndown": "^5.0.4"
}
}
```
--------------------------------------------------------------------------------
/index.ts:
--------------------------------------------------------------------------------
```typescript
#!/usr/bin/env node
// Core dependencies for MCP server and protocol handling
import { Server } from "@modelcontextprotocol/sdk/server/index.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import {
CallToolRequestSchema,
ListResourcesRequestSchema,
ListToolsRequestSchema,
ReadResourceRequestSchema,
ListPromptsRequestSchema,
GetPromptRequestSchema,
Tool,
Resource,
McpError,
ErrorCode,
TextContent,
ImageContent,
} from "@modelcontextprotocol/sdk/types.js";
// Web scraping and content processing dependencies
import { chromium, Browser, Page } from 'playwright';
import TurndownService from "turndown";
import type { Node } from "turndown";
import * as fs from 'fs';
import * as path from 'path';
import * as os from 'os';
// Add type declaration for window extensions
declare global {
interface Window {
chrome: {
runtime: Record<string, unknown>;
loadTimes: () => void;
csi: () => void;
app: Record<string, unknown>;
};
Notification: {
permission: NotificationPermission;
requestPermission?: (callback?: NotificationPermissionCallback) => Promise<NotificationPermission>;
};
}
}
// Initialize temp directory for screenshots
const SCREENSHOTS_DIR = fs.mkdtempSync(path.join(os.tmpdir(), 'mcp-screenshots-'));
// Initialize Turndown service for converting HTML to Markdown
// Configure with specific formatting preferences
const turndownService: TurndownService = new TurndownService({
headingStyle: 'atx', // Use # style headings
hr: '---', // Horizontal rule style
bulletListMarker: '-', // List item marker
codeBlockStyle: 'fenced', // Use ``` for code blocks
emDelimiter: '_', // Italics style
strongDelimiter: '**', // Bold style
linkStyle: 'inlined', // Use inline links
});
// Custom Turndown rules for better content extraction
// Remove script and style tags completely
turndownService.addRule('removeScripts', {
filter: ['script', 'style', 'noscript'],
replacement: () => ''
});
// Preserve link elements with their href attributes
turndownService.addRule('preserveLinks', {
filter: 'a',
replacement: (content: string, node: Node) => {
const element = node as HTMLAnchorElement;
const href = element.getAttribute('href');
return href ? `[${content}](${href})` : content;
}
});
// Preserve image elements with their src and alt attributes
turndownService.addRule('preserveImages', {
filter: 'img',
replacement: (content: string, node: Node) => {
const element = node as HTMLImageElement;
const alt = element.getAttribute('alt') || '';
const src = element.getAttribute('src');
return src ? `` : '';
}
});
// Core interfaces for research data management
interface ResearchResult {
url: string; // URL of the researched page
title: string; // Page title
content: string; // Extracted content in markdown
timestamp: string; // When the result was captured
screenshotPath?: string; // Path to screenshot file on disk
}
// Define structure for research session data
interface ResearchSession {
query: string; // Search query that initiated the session
results: ResearchResult[]; // Collection of research results
lastUpdated: string; // Timestamp of last update
}
// Screenshot management functions
async function saveScreenshot(screenshot: string, title: string): Promise<string> {
// Convert screenshot from base64 to buffer
const buffer = Buffer.from(screenshot, 'base64');
// Check size before saving
const MAX_SIZE = 5 * 1024 * 1024; // 5MB
if (buffer.length > MAX_SIZE) {
throw new McpError(
ErrorCode.InvalidRequest,
`Screenshot too large: ${Math.round(buffer.length / (1024 * 1024))}MB exceeds ${MAX_SIZE / (1024 * 1024)}MB limit`
);
}
// Generate a safe filename
const timestamp = new Date().getTime();
const safeTitle = title.replace(/[^a-z0-9]/gi, '_').toLowerCase();
const filename = `${safeTitle}-${timestamp}.png`;
const filepath = path.join(SCREENSHOTS_DIR, filename);
// Save the validated screenshot
await fs.promises.writeFile(filepath, buffer);
// Return the filepath to the saved screenshot
return filepath;
}
// Fix the cleanupScreenshots function
async function cleanupScreenshots(): Promise<void> {
try {
const files = await fs.promises.readdir(SCREENSHOTS_DIR);
await Promise.all(files.map((file: string) =>
fs.promises.unlink(path.join(SCREENSHOTS_DIR, file))
));
await fs.promises.rmdir(SCREENSHOTS_DIR);
} catch (error) {
console.error('Error cleaning up screenshots:', error);
}
}
// Available tools for web research functionality
const TOOLS: Tool[] = [
{
name: "search_google",
description: "Performs a web search using Google, ideal for finding current information, news, websites, and general knowledge. Use this tool when you need to research topics, find recent information, or gather data from the web. Returns structured search results with titles, URLs, and snippets.",
inputSchema: {
type: "object",
properties: {
query: { type: "string", description: "Search query" },
},
required: ["query"],
},
},
{
name: "visit_page",
description: "Navigates to a specific URL and extracts the page content in readable format, with option to capture a screenshot. Use this tool to deeply analyze specific web pages, read articles, examine documentation, or verify information directly from the source. Especially useful for in-depth research after identifying relevant pages via search.",
inputSchema: {
type: "object",
properties: {
url: { type: "string", description: "URL to visit" },
takeScreenshot: { type: "boolean", description: "Whether to take a screenshot" },
},
required: ["url"],
},
},
{
name: "take_screenshot",
description: "Captures a visual image of the currently loaded webpage. Use this tool when you need to preserve visual information, analyze page layouts, or document the current state of a webpage. Perfect for situations where textual content alone doesn't convey the full context.",
inputSchema: {
type: "object",
properties: {}, // No parameters needed
},
},
{
name: "search_scholar",
description: "Searches Google Scholar for academic papers and scholarly articles. Use this tool when researching scientific topics, looking for peer-reviewed research, academic citations, or scholarly literature. Returns structured data including titles, authors, publication details, and citation counts. Ideal for academic research and evidence-based inquiries.",
inputSchema: {
type: "object",
properties: {
query: { type: "string", description: "Academic search query" },
},
required: ["query"],
},
},
];
// Define available prompt types for type safety
type PromptName = "agentic-research";
// Define structure for research prompt arguments
interface AgenticResearchArgs {
topic: string; // Research topic provided by user
}
// Configure available prompts with their specifications
const PROMPTS = {
// Agentic research prompt configuration
"agentic-research": {
name: "agentic-research" as const, // Type-safe name
description: "Conduct iterative web research on a topic, exploring it thoroughly through multiple steps while maintaining a dialogue with the user",
arguments: [
{
name: "topic", // Topic argument specification
description: "The topic or question to research", // Description of the argument
required: true // Topic is mandatory
}
]
}
} as const; // Make object immutable
// Global state management for browser and research session
let browser: Browser | undefined; // Puppeteer browser instance
let page: Page | undefined; // Current active page
let currentSession: ResearchSession | undefined; // Current research session data
// Configuration constants for session management
const MAX_RESULTS_PER_SESSION = 100; // Maximum number of results to store per session
const MAX_RETRIES = 3; // Maximum retry attempts for operations
const RETRY_DELAY = 1000; // Delay between retries in milliseconds
// Generic retry mechanism for handling transient failures
async function withRetry<T>(
operation: () => Promise<T>, // Operation to retry
retries = MAX_RETRIES, // Number of retry attempts
delay = RETRY_DELAY // Delay between retries
): Promise<T> {
let lastError: Error;
// Attempt operation up to max retries
for (let i = 0; i < retries; i++) {
try {
return await operation();
} catch (error) {
lastError = error as Error;
if (i < retries - 1) {
console.error(`Attempt ${i + 1} failed, retrying in ${delay}ms:`, error);
await new Promise(resolve => setTimeout(resolve, delay));
}
}
}
throw lastError!; // Throw last error if all retries failed
}
// Add a new research result to the current session with data management
function addResult(result: ResearchResult): void {
// If no current session exists, initialize a new one
if (!currentSession) {
currentSession = {
query: "Research Session",
results: [],
lastUpdated: new Date().toISOString(),
};
}
// If the session has reached the maximum number of results, remove the oldest result
if (currentSession.results.length >= MAX_RESULTS_PER_SESSION) {
currentSession.results.shift();
}
// Add the new result to the session and update the last updated timestamp
currentSession.results.push(result);
currentSession.lastUpdated = new Date().toISOString();
}
// Safe page navigation with error handling and bot detection
async function safePageNavigation(page: Page, url: string): Promise<void> {
try {
// Step 1: Set cookies to bypass consent banner
await page.context().addCookies([{
name: 'CONSENT',
value: 'YES+',
domain: '.google.com',
path: '/'
}]);
// Step 2: Initial navigation
const response = await page.goto(url, {
waitUntil: 'domcontentloaded',
timeout: 15000
});
// Step 3: Basic response validation
if (!response) {
throw new Error('Navigation failed: no response received');
}
// Check HTTP status code; if 400 or higher, throw an error
const status = response.status();
if (status >= 400) {
throw new Error(`HTTP ${status}: ${response.statusText()}`);
}
// Step 4: Wait for network to become idle or timeout
await Promise.race([
page.waitForLoadState('networkidle', { timeout: 5000 })
.catch(() => {/* ignore timeout */ }),
// Fallback timeout in case networkidle never occurs
new Promise(resolve => setTimeout(resolve, 5000))
]);
// Step 5: Security and content validation
const validation = await page.evaluate(() => {
const botProtectionExists = [
'#challenge-running', // Cloudflare
'#cf-challenge-running', // Cloudflare
'#px-captcha', // PerimeterX
'#ddos-protection', // Various
'#waf-challenge-html' // Various WAFs
].some(selector => document.querySelector(selector));
// Check for suspicious page titles
const suspiciousTitle = [
'security check',
'ddos protection',
'please wait',
'just a moment',
'attention required'
].some(phrase => document.title.toLowerCase().includes(phrase));
// Count words in the page content
const bodyText = document.body.innerText || '';
const words = bodyText.trim().split(/\s+/).length;
// Return validation results
return {
wordCount: words,
botProtection: botProtectionExists,
suspiciousTitle,
title: document.title
};
});
// If bot protection is detected, throw an error
if (validation.botProtection) {
throw new Error('Bot protection detected');
}
// If the page title is suspicious, throw an error
if (validation.suspiciousTitle) {
throw new Error(`Suspicious page title detected: "${validation.title}"`);
}
// If the page contains insufficient content, throw an error
if (validation.wordCount < 1) {
throw new Error('Page contains insufficient content');
}
} catch (error) {
// If an error occurs during navigation, throw an error with the URL and the error message
throw new Error(`Navigation to ${url} failed: ${(error as Error).message}`);
}
}
// Take and optimize a screenshot
async function takeScreenshotWithSizeLimit(page: Page): Promise<string> {
const MAX_SIZE = 5 * 1024 * 1024;
const MAX_DIMENSION = 1920;
const MIN_DIMENSION = 800;
// Set viewport size
await page.setViewportSize({
width: 1920,
height: 1080
});
// Take initial screenshot
let screenshot = await page.screenshot({
type: 'png',
fullPage: false
});
// Handle buffer conversion
let buffer = screenshot;
let attempts = 0;
const MAX_ATTEMPTS = 3;
// While screenshot is too large, reduce size
while (buffer.length > MAX_SIZE && attempts < MAX_ATTEMPTS) {
// Get current viewport size
const viewport = page.viewportSize();
if (!viewport) continue;
// Calculate new dimensions
const scaleFactor = Math.pow(0.75, attempts + 1);
let newWidth = Math.round(viewport.width * scaleFactor);
let newHeight = Math.round(viewport.height * scaleFactor);
// Ensure dimensions are within bounds
newWidth = Math.max(MIN_DIMENSION, Math.min(MAX_DIMENSION, newWidth));
newHeight = Math.max(MIN_DIMENSION, Math.min(MAX_DIMENSION, newHeight));
// Update viewport with new dimensions
await page.setViewportSize({
width: newWidth,
height: newHeight
});
// Take new screenshot
screenshot = await page.screenshot({
type: 'png',
fullPage: false
});
// Update buffer with new screenshot
buffer = screenshot;
// Increment retry attempts
attempts++;
}
// Final attempt with minimum settings
if (buffer.length > MAX_SIZE) {
await page.setViewportSize({
width: MIN_DIMENSION,
height: MIN_DIMENSION
});
// Take final screenshot
screenshot = await page.screenshot({
type: 'png',
fullPage: false
});
// Update buffer with final screenshot
buffer = screenshot;
// Throw error if final screenshot is still too large
if (buffer.length > MAX_SIZE) {
throw new McpError(
ErrorCode.InvalidRequest,
`Failed to reduce screenshot to under 5MB even with minimum settings`
);
}
}
// Convert Buffer to base64 string before returning
return buffer.toString('base64');
}
// Initialize MCP server with basic configuration
const server: Server = new Server(
{
name: "webresearch", // Server name identifier
version: "0.1.6", // Server version number
},
{
capabilities: {
tools: {}, // Available tool configurations
resources: {}, // Resource handling capabilities
prompts: {} // Prompt processing capabilities
},
}
);
// Register handler for tool listing requests
server.setRequestHandler(ListToolsRequestSchema, async () => ({
tools: TOOLS // Return list of available research tools
}));
// Register handler for resource listing requests
server.setRequestHandler(ListResourcesRequestSchema, async () => {
// Return empty list if no active session
if (!currentSession) {
return { resources: [] };
}
// Compile list of available resources
const resources: Resource[] = [
// Add session summary resource
{
uri: "research://current/summary", // Resource identifier
name: "Current Research Session Summary",
description: "Summary of the current research session including queries and results",
mimeType: "application/json"
},
// Add screenshot resources if available
...currentSession.results
.map((r, i): Resource | undefined => r.screenshotPath ? {
uri: `research://screenshots/${i}`,
name: `Screenshot of ${r.title}`,
description: `Screenshot taken from ${r.url}`,
mimeType: "image/png"
} : undefined)
.filter((r): r is Resource => r !== undefined)
];
// Return compiled list of resources
return { resources };
});
// Register handler for resource content requests
server.setRequestHandler(ReadResourceRequestSchema, async (request) => {
const uri = request.params.uri.toString();
// Handle session summary requests for research data
if (uri === "research://current/summary") {
if (!currentSession) {
throw new McpError(
ErrorCode.InvalidRequest,
"No active research session"
);
}
// Return compiled list of resources
return {
contents: [{
uri,
mimeType: "application/json",
text: JSON.stringify({
query: currentSession.query,
resultCount: currentSession.results.length,
lastUpdated: currentSession.lastUpdated,
results: currentSession.results.map(r => ({
title: r.title,
url: r.url,
timestamp: r.timestamp,
screenshotPath: r.screenshotPath
}))
}, null, 2)
}]
};
}
// Handle screenshot requests
if (uri.startsWith("research://screenshots/")) {
const index = parseInt(uri.split("/").pop() || "", 10);
// Verify session exists
if (!currentSession) {
throw new McpError(
ErrorCode.InvalidRequest,
"No active research session"
);
}
// Verify index is within bounds
if (isNaN(index) || index < 0 || index >= currentSession.results.length) {
throw new McpError(
ErrorCode.InvalidRequest,
`Screenshot index out of bounds: ${index}`
);
}
// Get result containing screenshot
const result = currentSession.results[index];
if (!result?.screenshotPath) {
throw new McpError(
ErrorCode.InvalidRequest,
`No screenshot available at index: ${index}`
);
}
try {
// Read the binary data and convert to base64
const screenshotData = await fs.promises.readFile(result.screenshotPath);
// Convert Buffer to base64 string before returning
const base64Data = screenshotData.toString('base64');
// Return compiled list of resources
return {
contents: [{
uri,
mimeType: "image/png",
blob: base64Data
}]
};
} catch (error: unknown) {
// Handle error if screenshot cannot be read
const errorMessage = error instanceof Error ? error.message : 'Unknown error occurred';
throw new McpError(
ErrorCode.InternalError,
`Failed to read screenshot: ${errorMessage}`
);
}
}
// Handle unknown resource types
throw new McpError(
ErrorCode.InvalidRequest,
`Unknown resource: ${uri}`
);
});
// Initialize MCP server connection using stdio transport
const transport = new StdioServerTransport();
server.connect(transport).catch((error) => {
console.error("Failed to start server:", error);
process.exit(1);
});
// Convert HTML content to clean, readable markdown format
async function extractContentAsMarkdown(
page: Page, // Puppeteer page to extract from
selector?: string // Optional CSS selector to target specific content
): Promise<string> {
// Step 1: Execute content extraction in browser context
const html = await page.evaluate((sel) => {
// Handle case where specific selector is provided
if (sel) {
const element = document.querySelector(sel);
// Return element content or empty string if not found
return element ? element.outerHTML : '';
}
// Step 2: Try standard content containers first
const contentSelectors = [
'main', // HTML5 semantic main content
'article', // HTML5 semantic article content
'[role="main"]', // ARIA main content role
'#content', // Common content ID
'.content', // Common content class
'.main', // Alternative main class
'.post', // Blog post content
'.article', // Article content container
];
// Try each selector in priority order
for (const contentSelector of contentSelectors) {
const element = document.querySelector(contentSelector);
if (element) {
return element.outerHTML; // Return first matching content
}
}
// Step 3: Fallback to cleaning full body content
const body = document.body;
// Define elements to remove for cleaner content
const elementsToRemove = [
// Navigation elements
'header', // Page header
'footer', // Page footer
'nav', // Navigation sections
'[role="navigation"]', // ARIA navigation elements
// Sidebars and complementary content
'aside', // Sidebar content
'.sidebar', // Sidebar by class
'[role="complementary"]', // ARIA complementary content
// Navigation-related elements
'.nav', // Navigation classes
'.menu', // Menu elements
// Page structure elements
'.header', // Header classes
'.footer', // Footer classes
// Advertising and notices
'.advertisement', // Advertisement containers
'.ads', // Ad containers
'.cookie-notice', // Cookie consent notices
];
// Remove each unwanted element from content
elementsToRemove.forEach(sel => {
body.querySelectorAll(sel).forEach(el => el.remove());
});
// Return cleaned body content
return body.outerHTML;
}, selector);
// Step 4: Handle empty content case
if (!html) {
return '';
}
try {
// Step 5: Convert HTML to Markdown
const markdown = turndownService.turndown(html);
// Step 6: Clean up and format markdown
return markdown
.replace(/\n{3,}/g, '\n\n') // Replace excessive newlines with double
.replace(/^- $/gm, '') // Remove empty list items
.replace(/^\s+$/gm, '') // Remove whitespace-only lines
.trim(); // Remove leading/trailing whitespace
} catch (error) {
// Log conversion errors and return original HTML as fallback
console.error('Error converting HTML to Markdown:', error);
return html;
}
}
// Validate URL format and ensure security constraints
function isValidUrl(urlString: string): boolean {
try {
// Attempt to parse URL string
const url = new URL(urlString);
// Only allow HTTP and HTTPS protocols for security
return url.protocol === 'http:' || url.protocol === 'https:';
} catch {
// Return false for any invalid URL format
return false;
}
}
// Define result type for tool operations
type ToolResult = {
content: (TextContent | ImageContent)[]; // Array of text or image content
isError?: boolean; // Optional error flag
};
// Tool request handler for executing research operations
server.setRequestHandler(CallToolRequestSchema, async (request): Promise<ToolResult> => {
// Initialize browser for tool operations
const page = await ensureBrowser();
switch (request.params.name) {
// Handle Google search operations
case "search_google": {
// Extract search query from request parameters
const { query } = request.params.arguments as { query: string };
try {
// Execute search with retry mechanism
const results = await withRetry(async () => {
// Step 1: Navigate to Google search page
await safePageNavigation(page, 'https://www.google.com');
// Step 2: Find and interact with search input
await withRetry(async () => {
// Wait for any search input element to appear
await Promise.race([
// Try multiple possible selectors for search input
page.waitForSelector('input[name="q"]', { timeout: 5000 }),
page.waitForSelector('textarea[name="q"]', { timeout: 5000 }),
page.waitForSelector('input[type="text"]', { timeout: 5000 })
]).catch(() => {
throw new Error('Search input not found - no matching selectors');
});
// Find the actual search input element
const searchInput = await page.$('input[name="q"]') ||
await page.$('textarea[name="q"]') ||
await page.$('input[type="text"]');
// Verify search input was found
if (!searchInput) {
throw new Error('Search input element not found after waiting');
}
// Step 3: Enter search query
await searchInput.click({ clickCount: 3 }); // Select all existing text
await searchInput.press('Backspace'); // Clear selected text
await searchInput.type(query); // Type new query
}, 3, 2000); // Allow 3 retries with 2s delay
// Step 4: Submit search and wait for results
await withRetry(async () => {
await Promise.all([
page.keyboard.press('Enter'),
page.waitForLoadState('networkidle', { timeout: 15000 }),
]);
});
// Step 5: Extract search results
const searchResults = await withRetry(async () => {
const results = await page.evaluate(() => {
// Find all search result containers
const elements = document.querySelectorAll('div.g');
if (!elements || elements.length === 0) {
throw new Error('No search results found');
}
// Extract data from each result
return Array.from(elements).map((el) => {
// Find required elements within result container
const titleEl = el.querySelector('h3'); // Title element
const linkEl = el.querySelector('a'); // Link element
const snippetEl = el.querySelector('div.VwiC3b'); // Snippet element
// Skip results missing required elements
if (!titleEl || !linkEl || !snippetEl) {
return null;
}
// Return structured result data
return {
title: titleEl.textContent || '', // Result title
url: linkEl.getAttribute('href') || '', // Result URL
snippet: snippetEl.textContent || '', // Result description
};
}).filter(result => result !== null); // Remove invalid results
});
// Verify we found valid results
if (!results || results.length === 0) {
throw new Error('No valid search results found');
}
// Return compiled list of results
return results;
});
// Step 6: Store results in session
searchResults.forEach((result) => {
addResult({
url: result.url,
title: result.title,
content: result.snippet,
timestamp: new Date().toISOString(),
});
});
// Return compiled list of results
return searchResults;
});
// Step 7: Return formatted results
return {
content: [{
type: "text",
text: JSON.stringify(results, null, 2) // Pretty-print JSON results
}]
};
} catch (error) {
// Handle and format search errors
return {
content: [{
type: "text",
text: `Failed to perform search: ${(error as Error).message}`
}],
isError: true
};
}
}
// Handle Google Scholar search operations
case "search_scholar": {
// Extract search query from request parameters
const { query } = request.params.arguments as { query: string };
try {
// Execute search with retry mechanism
const results = await withRetry(async () => {
// Step 1: Navigate to Google Scholar search page
await safePageNavigation(page, 'https://scholar.google.com');
// Step 2: Find and interact with search input
await withRetry(async () => {
// Wait for search input element to appear
await page.waitForSelector('input[name="q"]', { timeout: 5000 })
.catch(() => {
throw new Error('Scholar search input not found');
});
// Find the search input element
const searchInput = await page.$('input[name="q"]');
// Verify search input was found
if (!searchInput) {
throw new Error('Scholar search input element not found after waiting');
}
// Step 3: Enter search query
await searchInput.click({ clickCount: 3 }); // Select all existing text
await searchInput.press('Backspace'); // Clear selected text
await searchInput.type(query); // Type new query
}, 3, 2000); // Allow 3 retries with 2s delay
// Step 4: Submit search and wait for results
await withRetry(async () => {
await Promise.all([
page.keyboard.press('Enter'),
page.waitForLoadState('networkidle', { timeout: 15000 }),
]);
});
// Step 5: Extract scholar search results
const scholarResults = await withRetry(async () => {
const results = await page.evaluate(() => {
// Find all scholar result containers
const elements = document.querySelectorAll('.gs_r.gs_or.gs_scl');
if (!elements || elements.length === 0) {
throw new Error('No scholar search results found');
}
// Extract data from each result
return Array.from(elements).map((el) => {
try {
// Find required elements within result container
const titleEl = el.querySelector('.gs_rt'); // Title element
const authorEl = el.querySelector('.gs_a'); // Authors, venue, year
const snippetEl = el.querySelector('.gs_rs'); // Snippet/abstract
const citedByEl = el.querySelector('.gs_fl a:nth-child(3)'); // Cited by element
// Extract title and URL
let title = '';
let url = '';
if (titleEl) {
const titleLink = titleEl.querySelector('a');
title = titleEl.textContent?.trim() || '';
url = titleLink?.getAttribute('href') || '';
}
// Extract author, venue, and year information
const authorInfo = authorEl?.textContent?.trim() || '';
// Extract snippet
const snippet = snippetEl?.textContent?.trim() || '';
// Extract citation count
let citationCount = '';
if (citedByEl && citedByEl.textContent?.includes('Cited by')) {
citationCount = citedByEl.textContent.trim();
}
// Skip results missing critical data
if (!title) {
return null;
}
// Return structured result data
return {
title, // Paper title
url, // Paper URL if available
authorInfo, // Authors, venue, year
snippet, // Abstract/snippet
citationCount, // Citation information
};
} catch (err) {
// Skip problematic results
return null;
}
}).filter(result => result !== null); // Remove invalid results
});
// Verify we found valid results
if (!results || results.length === 0) {
throw new Error('No valid scholar search results found');
}
// Return compiled list of results
return results;
});
// Step 6: Store results in session
scholarResults.forEach((result) => {
addResult({
url: result.url || 'https://scholar.google.com',
title: result.title,
content: `${result.authorInfo}\n\n${result.snippet}\n\n${result.citationCount}`,
timestamp: new Date().toISOString(),
});
});
// Return compiled list of results
return scholarResults;
});
// Step 7: Return formatted results
return {
content: [{
type: "text",
text: JSON.stringify(results, null, 2) // Pretty-print JSON results
}]
};
} catch (error) {
// Handle and format search errors
return {
content: [{
type: "text",
text: `Failed to perform scholar search: ${(error as Error).message}`
}],
isError: true
};
}
}
// Handle webpage visit and content extraction
case "visit_page": {
// Extract URL and screenshot flag from request
const { url, takeScreenshot } = request.params.arguments as {
url: string; // Target URL to visit
takeScreenshot?: boolean; // Optional screenshot flag
};
// Step 1: Validate URL format and security
if (!isValidUrl(url)) {
return {
content: [{
type: "text" as const,
text: `Invalid URL: ${url}. Only http and https protocols are supported.`
}],
isError: true
};
}
try {
// Step 2: Visit page and extract content with retry mechanism
const result = await withRetry(async () => {
// Navigate to target URL safely
await safePageNavigation(page, url);
const title = await page.title();
// Step 3: Extract and process page content
const content = await withRetry(async () => {
// Convert page content to markdown
const extractedContent = await extractContentAsMarkdown(page);
// If no content is extracted, throw an error
if (!extractedContent) {
throw new Error('Failed to extract content');
}
// Return the extracted content
return extractedContent;
});
// Step 4: Create result object with page data
const pageResult: ResearchResult = {
url, // Original URL
title, // Page title
content, // Markdown content
timestamp: new Date().toISOString(), // Capture time
};
// Step 5: Take screenshot if requested
let screenshotUri: string | undefined;
if (takeScreenshot) {
// Capture and process screenshot
const screenshot = await takeScreenshotWithSizeLimit(page);
pageResult.screenshotPath = await saveScreenshot(screenshot, title);
// Get the index for the resource URI
const resultIndex = currentSession ? currentSession.results.length : 0;
screenshotUri = `research://screenshots/${resultIndex}`;
// Notify clients about new screenshot resource
server.notification({
method: "notifications/resources/list_changed"
});
}
// Step 6: Store result in session
addResult(pageResult);
return { pageResult, screenshotUri };
});
// Step 7: Return formatted result with screenshot URI if taken
const response: ToolResult = {
content: [{
type: "text" as const,
text: JSON.stringify({
url: result.pageResult.url,
title: result.pageResult.title,
content: result.pageResult.content,
timestamp: result.pageResult.timestamp,
screenshot: result.screenshotUri ? `View screenshot via *MCP Resources* (Paperclip icon) @ URI: ${result.screenshotUri}` : undefined
}, null, 2)
}]
};
return response;
} catch (error) {
// Handle and format page visit errors
return {
content: [{
type: "text" as const,
text: `Failed to visit page: ${(error as Error).message}`
}],
isError: true
};
}
}
// Handle standalone screenshot requests
case "take_screenshot": {
try {
// Step 1: Capture screenshot with retry mechanism
const screenshot = await withRetry(async () => {
// Take and optimize screenshot with default size limits
return await takeScreenshotWithSizeLimit(page);
});
// Step 2: Initialize session if needed
if (!currentSession) {
currentSession = {
query: "Screenshot Session", // Session identifier
results: [], // Empty results array
lastUpdated: new Date().toISOString(), // Current timestamp
};
}
// Step 3: Get current page information
const pageUrl = await page.url(); // Current page URL
const pageTitle = await page.title(); // Current page title
// Step 4: Save screenshot to disk
const screenshotPath = await saveScreenshot(screenshot, pageTitle || 'untitled');
// Step 5: Create and store screenshot result
const resultIndex = currentSession ? currentSession.results.length : 0;
addResult({
url: pageUrl,
title: pageTitle || "Untitled Page", // Fallback title if none available
content: "Screenshot taken", // Simple content description
timestamp: new Date().toISOString(), // Capture time
screenshotPath // Path to screenshot file
});
// Step 6: Notify clients about new screenshot resource
server.notification({
method: "notifications/resources/list_changed"
});
// Step 7: Return success message with resource URI
const resourceUri = `research://screenshots/${resultIndex}`;
return {
content: [{
type: "text" as const,
text: `Screenshot taken successfully. You can view it via *MCP Resources* (Paperclip icon) @ URI: ${resourceUri}`
}]
};
} catch (error) {
// Handle and format screenshot errors
return {
content: [{
type: "text" as const,
text: `Failed to take screenshot: ${(error as Error).message}`
}],
isError: true
};
}
}
// Handle unknown tool requests
default:
throw new McpError(
ErrorCode.MethodNotFound,
`Unknown tool: ${request.params.name}`
);
}
});
// Register handler for prompt listing requests
server.setRequestHandler(ListPromptsRequestSchema, async () => {
// Return all available prompts
return { prompts: Object.values(PROMPTS) };
});
// Register handler for prompt retrieval and execution
server.setRequestHandler(GetPromptRequestSchema, async (request) => {
// Extract and validate prompt name
const promptName = request.params.name as PromptName;
const prompt = PROMPTS[promptName];
// Handle unknown prompt requests
if (!prompt) {
throw new McpError(ErrorCode.InvalidRequest, `Prompt not found: ${promptName}`);
}
// Handle agentic research prompt
if (promptName === "agentic-research") {
// Extract research topic from request arguments
const args = request.params.arguments as AgenticResearchArgs | undefined;
const topic = args?.topic || ""; // Use empty string if no topic provided
// Return research assistant prompt with instructions
return {
messages: [
// Initial assistant message establishing role
{
role: "assistant",
content: {
type: "text",
text: "I am ready to help you with your research. I will conduct thorough web research, explore topics deeply, and maintain a dialogue with you throughout the process."
}
},
// Detailed research instructions for the user
{
role: "user",
content: {
type: "text",
text: `I'd like to research this topic: <topic>${topic}</topic>
Please help me explore it deeply, like you're a thoughtful, highly-trained research assistant.
General instructions:
1. Start by proposing your research approach -- namely, formulate what initial query you will use to search the web. Propose a relatively broad search to understand the topic landscape. At the same time, make your queries optimized for returning high-quality results based on what you know about constructing Google search queries.
2. Next, get my input on whether you should proceed with that query or if you should refine it.
3. Once you have an approved query, perform the search.
4. Prioritize high quality, authoritative sources when they are available and relevant to the topic. Avoid low quality or spammy sources.
5. Retrieve information that is relevant to the topic at hand.
6. Iteratively refine your research direction based on what you find.
7. Keep me informed of what you find and let *me* guide the direction of the research interactively.
8. If you run into a dead end while researching, do a Google search for the topic and attempt to find a URL for a relevant page. Then, explore that page in depth.
9. Only conclude when my research goals are met.
10. **Always cite your sources**, providing URLs to the sources you used in a citation block at the end of your response.
You can use these tools:
- search_google: Search for information
- visit_page: Visit and extract content from web pages
Do *NOT* use the following tools:
- Anything related to knowledge graphs or memory, unless explicitly instructed to do so by the user.`
}
}
]
};
}
// Handle unsupported prompt types
throw new McpError(ErrorCode.InvalidRequest, "Prompt implementation not found");
});
// In the ensureBrowser function, modify the initialization script:
async function ensureBrowser(): Promise<Page> {
if (!browser) {
browser = await chromium.launch({
headless: true,
args: [
'--disable-blink-features=AutomationControlled',
'--disable-features=IsolateOrigins,site-per-process',
'--disable-site-isolation-trials',
'--disable-setuid-sandbox',
'--no-sandbox',
'--disable-dev-shm-usage',
'--disable-accelerated-2d-canvas',
'--no-first-run',
'--no-service-autorun',
'--password-store=basic',
'--system-developer-mode',
'--enable-javascript',
`--window-size=${1366 + Math.floor(Math.random() * 100)},${768 + Math.floor(Math.random() * 100)}`,
]
});
const context = await browser.newContext({
userAgent: 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36',
viewport: { width: 1366, height: 768 }, // Set fixed viewport instead of null
deviceScaleFactor: 1,
javaScriptEnabled: true,
});
await context.addInitScript(() => {
Object.defineProperty(navigator, 'webdriver', {
get: () => undefined
});
Object.defineProperty(navigator, 'permissions', {
get: () => ({
query: async () => ({ state: 'prompt' as PermissionState })
})
});
window.chrome = {
runtime: {},
loadTimes: function(){},
csi: function(){},
app: {},
};
const originalToString = Error.prototype.toString;
Error.prototype.toString = function(this: Error) {
return originalToString.call(this).replace(/\n.*puppeteer.*\n/g, '\n');
};
if (!window.Notification) {
const NotificationClass = function(title: string, options?: NotificationOptions) {
return {
title,
...options
};
} as unknown as typeof Notification;
Object.defineProperty(NotificationClass, 'permission', {
value: 'default' as NotificationPermission,
writable: false,
configurable: false,
enumerable: true
});
NotificationClass.requestPermission = async () => 'default' as NotificationPermission;
NotificationClass.prototype = {} as Notification;
Object.defineProperty(window, 'Notification', {
value: NotificationClass,
writable: false,
configurable: false
});
}
Object.defineProperty(navigator, 'languages', {
get: () => ['en-US', 'en']
});
});
page = await context.newPage();
await page.route('**', async (route) => {
const request = route.request();
if (request.resourceType() === 'script') {
route.continue({
headers: {
...request.headers(),
'sec-ch-ua': '"Not_A Brand";v="8", "Chromium";v="120", "Google Chrome";v="120"',
'sec-ch-ua-mobile': '?0',
'sec-ch-ua-platform': '"Windows"'
}
});
} else {
route.continue();
}
});
}
if (!page) {
const context = await browser.newContext();
page = await context.newPage();
}
return page;
}
// Cleanup function
async function cleanup(): Promise<void> {
try {
// Clean up screenshots first
await cleanupScreenshots();
// Then close the browser
if (browser) {
await browser.close();
}
} catch (error) {
console.error('Error during cleanup:', error);
} finally {
browser = undefined;
page = undefined;
}
}
// Register cleanup handlers
process.on('exit', cleanup);
process.on('SIGTERM', cleanup);
process.on('SIGINT', cleanup);
process.on('SIGHUP', cleanup);
```
--------------------------------------------------------------------------------
/docs/mcp_spec/llms-full.txt:
--------------------------------------------------------------------------------
```
# Clients
A list of applications that support MCP integrations
This page provides an overview of applications that support the Model Context Protocol (MCP). Each client may support different MCP features, allowing for varying levels of integration with MCP servers.
## Feature support matrix
| Client | [Resources] | [Prompts] | [Tools] | [Sampling] | Roots | Notes |
| ---------------------------- | ----------- | --------- | ------- | ---------- | ----- | ------------------------------------------------ |
| [Claude Desktop App][Claude] | ✅ | ✅ | ✅ | ❌ | ❌ | Full support for all MCP features |
| [Zed][Zed] | ❌ | ✅ | ❌ | ❌ | ❌ | Prompts appear as slash commands |
| [Sourcegraph Cody][Cody] | ✅ | ❌ | ❌ | ❌ | ❌ | Supports resources through OpenCTX |
| [Firebase Genkit][Genkit] | ⚠️ | ✅ | ✅ | ❌ | ❌ | Supports resource list and lookup through tools. |
| [Continue][Continue] | ✅ | ✅ | ✅ | ❌ | ❌ | Full support for all MCP features |
[Claude]: https://claude.ai/download
[Zed]: https://zed.dev
[Cody]: https://sourcegraph.com/cody
[Genkit]: https://github.com/firebase/genkit
[Continue]: https://github.com/continuedev/continue
[Resources]: https://modelcontextprotocol.io/docs/concepts/resources
[Prompts]: https://modelcontextprotocol.io/docs/concepts/prompts
[Tools]: https://modelcontextprotocol.io/docs/concepts/tools
[Sampling]: https://modelcontextprotocol.io/docs/concepts/sampling
## Client details
### Claude Desktop App
The Claude desktop application provides comprehensive support for MCP, enabling deep integration with local tools and data sources.
**Key features:**
* Full support for resources, allowing attachment of local files and data
* Support for prompt templates
* Tool integration for executing commands and scripts
* Local server connections for enhanced privacy and security
> ⓘ Note: The Claude.ai web application does not currently support MCP. MCP features are only available in the desktop application.
### Zed
[Zed](https://zed.dev/docs/assistant/model-context-protocol) is a high-performance code editor with built-in MCP support, focusing on prompt templates and tool integration.
**Key features:**
* Prompt templates surface as slash commands in the editor
* Tool integration for enhanced coding workflows
* Tight integration with editor features and workspace context
* Does not support MCP resources
### Sourcegraph Cody
[Cody](https://openctx.org/docs/providers/modelcontextprotocol) is Sourcegraph's AI coding assistant, which implements MCP through OpenCTX.
**Key features:**
* Support for MCP resources
* Integration with Sourcegraph's code intelligence
* Uses OpenCTX as an abstraction layer
* Future support planned for additional MCP features
### Firebase Genkit
[Genkit](https://github.com/firebase/genkit) is Firebase's SDK for building and integrating GenAI features into applications. The [genkitx-mcp](https://github.com/firebase/genkit/tree/main/js/plugins/mcp) plugin enables consuming MCP servers as a client or creating MCP servers from Genkit tools and prompts.
**Key features:**
* Client support for tools and prompts (resources partially supported)
* Rich discovery with support in Genkit's Dev UI playground
* Seamless interoperability with Genkit's existing tools and prompts
* Works across a wide variety of GenAI models from top providers
### Continue
[Continue](https://github.com/continuedev/continue) is an open-source AI code assistant, with built-in support for all MCP features.
**Key features**
* Type "@" to mention MCP resources
* Prompt templates surface as slash commands
* Use both built-in and MCP tools directly in chat
* Supports VS Code and JetBrains IDEs, with any LLM
## Adding MCP support to your application
If you've added MCP support to your application, we encourage you to submit a pull request to add it to this list. MCP integration can provide your users with powerful contextual AI capabilities and make your application part of the growing MCP ecosystem.
Benefits of adding MCP support:
* Enable users to bring their own context and tools
* Join a growing ecosystem of interoperable AI applications
* Provide users with flexible integration options
* Support local-first AI workflows
To get started with implementing MCP in your application, check out our [Python](https://github.com/modelcontextprotocol/python-sdk) or [TypeScript SDK Documentation](https://github.com/modelcontextprotocol/typescript-sdk)
## Updates and corrections
This list is maintained by the community. If you notice any inaccuracies or would like to update information about MCP support in your application, please submit a pull request or [open an issue in our documentation repository](https://github.com/modelcontextprotocol/docs/issues).
# Core architecture
Understand how MCP connects clients, servers, and LLMs
The Model Context Protocol (MCP) is built on a flexible, extensible architecture that enables seamless communication between LLM applications and integrations. This document covers the core architectural components and concepts.
## Overview
MCP follows a client-server architecture where:
* **Hosts** are LLM applications (like Claude Desktop or IDEs) that initiate connections
* **Clients** maintain 1:1 connections with servers, inside the host application
* **Servers** provide context, tools, and prompts to clients
```mermaid
flowchart LR
subgraph " Host (e.g., Claude Desktop) "
client1[MCP Client]
client2[MCP Client]
end
subgraph "Server Process"
server1[MCP Server]
end
subgraph "Server Process"
server2[MCP Server]
end
client1 <-->|Transport Layer| server1
client2 <-->|Transport Layer| server2
```
## Core components
### Protocol layer
The protocol layer handles message framing, request/response linking, and high-level communication patterns.
<Tabs>
<Tab title="TypeScript">
```typescript
class Protocol<Request, Notification, Result> {
// Handle incoming requests
setRequestHandler<T>(schema: T, handler: (request: T, extra: RequestHandlerExtra) => Promise<Result>): void
// Handle incoming notifications
setNotificationHandler<T>(schema: T, handler: (notification: T) => Promise<void>): void
// Send requests and await responses
request<T>(request: Request, schema: T, options?: RequestOptions): Promise<T>
// Send one-way notifications
notification(notification: Notification): Promise<void>
}
```
</Tab>
<Tab title="Python">
```python
class Session(BaseSession[RequestT, NotificationT, ResultT]):
async def send_request(
self,
request: RequestT,
result_type: type[Result]
) -> Result:
"""
Send request and wait for response. Raises McpError if response contains error.
"""
# Request handling implementation
async def send_notification(
self,
notification: NotificationT
) -> None:
"""Send one-way notification that doesn't expect response."""
# Notification handling implementation
async def _received_request(
self,
responder: RequestResponder[ReceiveRequestT, ResultT]
) -> None:
"""Handle incoming request from other side."""
# Request handling implementation
async def _received_notification(
self,
notification: ReceiveNotificationT
) -> None:
"""Handle incoming notification from other side."""
# Notification handling implementation
```
</Tab>
</Tabs>
Key classes include:
* `Protocol`
* `Client`
* `Server`
### Transport layer
The transport layer handles the actual communication between clients and servers. MCP supports multiple transport mechanisms:
1. **Stdio transport**
* Uses standard input/output for communication
* Ideal for local processes
2. **HTTP with SSE transport**
* Uses Server-Sent Events for server-to-client messages
* HTTP POST for client-to-server messages
All transports use [JSON-RPC](https://www.jsonrpc.org/) 2.0 to exchange messages. See the [specification](https://spec.modelcontextprotocol.io) for detailed information about the Model Context Protocol message format.
### Message types
MCP has these main types of messages:
1. **Requests** expect a response from the other side:
```typescript
interface Request {
method: string;
params?: { ... };
}
```
2. **Notifications** are one-way messages that don't expect a response:
```typescript
interface Notification {
method: string;
params?: { ... };
}
```
3. **Results** are successful responses to requests:
```typescript
interface Result {
[key: string]: unknown;
}
```
4. **Errors** indicate that a request failed:
```typescript
interface Error {
code: number;
message: string;
data?: unknown;
}
```
## Connection lifecycle
### 1. Initialization
```mermaid
sequenceDiagram
participant Client
participant Server
Client->>Server: initialize request
Server->>Client: initialize response
Client->>Server: initialized notification
Note over Client,Server: Connection ready for use
```
1. Client sends `initialize` request with protocol version and capabilities
2. Server responds with its protocol version and capabilities
3. Client sends `initialized` notification as acknowledgment
4. Normal message exchange begins
### 2. Message exchange
After initialization, the following patterns are supported:
* **Request-Response**: Client or server sends requests, the other responds
* **Notifications**: Either party sends one-way messages
### 3. Termination
Either party can terminate the connection:
* Clean shutdown via `close()`
* Transport disconnection
* Error conditions
## Error handling
MCP defines these standard error codes:
```typescript
enum ErrorCode {
// Standard JSON-RPC error codes
ParseError = -32700,
InvalidRequest = -32600,
MethodNotFound = -32601,
InvalidParams = -32602,
InternalError = -32603
}
```
SDKs and applications can define their own error codes above -32000.
Errors are propagated through:
* Error responses to requests
* Error events on transports
* Protocol-level error handlers
## Implementation example
Here's a basic example of implementing an MCP server:
<Tabs>
<Tab title="TypeScript">
```typescript
import { Server } from "@modelcontextprotocol/sdk/server/index.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
const server = new Server({
name: "example-server",
version: "1.0.0"
}, {
capabilities: {
resources: {}
}
});
// Handle requests
server.setRequestHandler(ListResourcesRequestSchema, async () => {
return {
resources: [
{
uri: "example://resource",
name: "Example Resource"
}
]
};
});
// Connect transport
const transport = new StdioServerTransport();
await server.connect(transport);
```
</Tab>
<Tab title="Python">
```python
import asyncio
import mcp.types as types
from mcp.server import Server
from mcp.server.stdio import stdio_server
app = Server("example-server")
@app.list_resources()
async def list_resources() -> list[types.Resource]:
return [
types.Resource(
uri="example://resource",
name="Example Resource"
)
]
async def main():
async with stdio_server() as streams:
await app.run(
streams[0],
streams[1],
app.create_initialization_options()
)
if __name__ == "__main__":
asyncio.run(main)
```
</Tab>
</Tabs>
## Best practices
### Transport selection
1. **Local communication**
* Use stdio transport for local processes
* Efficient for same-machine communication
* Simple process management
2. **Remote communication**
* Use SSE for scenarios requiring HTTP compatibility
* Consider security implications including authentication and authorization
### Message handling
1. **Request processing**
* Validate inputs thoroughly
* Use type-safe schemas
* Handle errors gracefully
* Implement timeouts
2. **Progress reporting**
* Use progress tokens for long operations
* Report progress incrementally
* Include total progress when known
3. **Error management**
* Use appropriate error codes
* Include helpful error messages
* Clean up resources on errors
## Security considerations
1. **Transport security**
* Use TLS for remote connections
* Validate connection origins
* Implement authentication when needed
2. **Message validation**
* Validate all incoming messages
* Sanitize inputs
* Check message size limits
* Verify JSON-RPC format
3. **Resource protection**
* Implement access controls
* Validate resource paths
* Monitor resource usage
* Rate limit requests
4. **Error handling**
* Don't leak sensitive information
* Log security-relevant errors
* Implement proper cleanup
* Handle DoS scenarios
## Debugging and monitoring
1. **Logging**
* Log protocol events
* Track message flow
* Monitor performance
* Record errors
2. **Diagnostics**
* Implement health checks
* Monitor connection state
* Track resource usage
* Profile performance
3. **Testing**
* Test different transports
* Verify error handling
* Check edge cases
* Load test servers
# Prompts
Create reusable prompt templates and workflows
Prompts enable servers to define reusable prompt templates and workflows that clients can easily surface to users and LLMs. They provide a powerful way to standardize and share common LLM interactions.
<Note>
Prompts are designed to be **user-controlled**, meaning they are exposed from servers to clients with the intention of the user being able to explicitly select them for use.
</Note>
## Overview
Prompts in MCP are predefined templates that can:
* Accept dynamic arguments
* Include context from resources
* Chain multiple interactions
* Guide specific workflows
* Surface as UI elements (like slash commands)
## Prompt structure
Each prompt is defined with:
```typescript
{
name: string; // Unique identifier for the prompt
description?: string; // Human-readable description
arguments?: [ // Optional list of arguments
{
name: string; // Argument identifier
description?: string; // Argument description
required?: boolean; // Whether argument is required
}
]
}
```
## Discovering prompts
Clients can discover available prompts through the `prompts/list` endpoint:
```typescript
// Request
{
method: "prompts/list"
}
// Response
{
prompts: [
{
name: "analyze-code",
description: "Analyze code for potential improvements",
arguments: [
{
name: "language",
description: "Programming language",
required: true
}
]
}
]
}
```
## Using prompts
To use a prompt, clients make a `prompts/get` request:
````typescript
// Request
{
method: "prompts/get",
params: {
name: "analyze-code",
arguments: {
language: "python"
}
}
}
// Response
{
description: "Analyze Python code for potential improvements",
messages: [
{
role: "user",
content: {
type: "text",
text: "Please analyze the following Python code for potential improvements:\n\n```python\ndef calculate_sum(numbers):\n total = 0\n for num in numbers:\n total = total + num\n return total\n\nresult = calculate_sum([1, 2, 3, 4, 5])\nprint(result)\n```"
}
}
]
}
````
## Dynamic prompts
Prompts can be dynamic and include:
### Embedded resource context
```json
{
"name": "analyze-project",
"description": "Analyze project logs and code",
"arguments": [
{
"name": "timeframe",
"description": "Time period to analyze logs",
"required": true
},
{
"name": "fileUri",
"description": "URI of code file to review",
"required": true
}
]
}
```
When handling the `prompts/get` request:
```json
{
"messages": [
{
"role": "user",
"content": {
"type": "text",
"text": "Analyze these system logs and the code file for any issues:"
}
},
{
"role": "user",
"content": {
"type": "resource",
"resource": {
"uri": "logs://recent?timeframe=1h",
"text": "[2024-03-14 15:32:11] ERROR: Connection timeout in network.py:127\n[2024-03-14 15:32:15] WARN: Retrying connection (attempt 2/3)\n[2024-03-14 15:32:20] ERROR: Max retries exceeded",
"mimeType": "text/plain"
}
}
},
{
"role": "user",
"content": {
"type": "resource",
"resource": {
"uri": "file:///path/to/code.py",
"text": "def connect_to_service(timeout=30):\n retries = 3\n for attempt in range(retries):\n try:\n return establish_connection(timeout)\n except TimeoutError:\n if attempt == retries - 1:\n raise\n time.sleep(5)\n\ndef establish_connection(timeout):\n # Connection implementation\n pass",
"mimeType": "text/x-python"
}
}
}
]
}
```
### Multi-step workflows
```typescript
const debugWorkflow = {
name: "debug-error",
async getMessages(error: string) {
return [
{
role: "user",
content: {
type: "text",
text: `Here's an error I'm seeing: ${error}`
}
},
{
role: "assistant",
content: {
type: "text",
text: "I'll help analyze this error. What have you tried so far?"
}
},
{
role: "user",
content: {
type: "text",
text: "I've tried restarting the service, but the error persists."
}
}
];
}
};
```
## Example implementation
Here's a complete example of implementing prompts in an MCP server:
<Tabs>
<Tab title="TypeScript">
```typescript
import { Server } from "@modelcontextprotocol/sdk/server";
import {
ListPromptsRequestSchema,
GetPromptRequestSchema
} from "@modelcontextprotocol/sdk/types";
const PROMPTS = {
"git-commit": {
name: "git-commit",
description: "Generate a Git commit message",
arguments: [
{
name: "changes",
description: "Git diff or description of changes",
required: true
}
]
},
"explain-code": {
name: "explain-code",
description: "Explain how code works",
arguments: [
{
name: "code",
description: "Code to explain",
required: true
},
{
name: "language",
description: "Programming language",
required: false
}
]
}
};
const server = new Server({
name: "example-prompts-server",
version: "1.0.0"
}, {
capabilities: {
prompts: {}
}
});
// List available prompts
server.setRequestHandler(ListPromptsRequestSchema, async () => {
return {
prompts: Object.values(PROMPTS)
};
});
// Get specific prompt
server.setRequestHandler(GetPromptRequestSchema, async (request) => {
const prompt = PROMPTS[request.params.name];
if (!prompt) {
throw new Error(`Prompt not found: ${request.params.name}`);
}
if (request.params.name === "git-commit") {
return {
messages: [
{
role: "user",
content: {
type: "text",
text: `Generate a concise but descriptive commit message for these changes:\n\n${request.params.arguments?.changes}`
}
}
]
};
}
if (request.params.name === "explain-code") {
const language = request.params.arguments?.language || "Unknown";
return {
messages: [
{
role: "user",
content: {
type: "text",
text: `Explain how this ${language} code works:\n\n${request.params.arguments?.code}`
}
}
]
};
}
throw new Error("Prompt implementation not found");
});
```
</Tab>
<Tab title="Python">
```python
from mcp.server import Server
import mcp.types as types
# Define available prompts
PROMPTS = {
"git-commit": types.Prompt(
name="git-commit",
description="Generate a Git commit message",
arguments=[
types.PromptArgument(
name="changes",
description="Git diff or description of changes",
required=True
)
],
),
"explain-code": types.Prompt(
name="explain-code",
description="Explain how code works",
arguments=[
types.PromptArgument(
name="code",
description="Code to explain",
required=True
),
types.PromptArgument(
name="language",
description="Programming language",
required=False
)
],
)
}
# Initialize server
app = Server("example-prompts-server")
@app.list_prompts()
async def list_prompts() -> list[types.Prompt]:
return list(PROMPTS.values())
@app.get_prompt()
async def get_prompt(
name: str, arguments: dict[str, str] | None = None
) -> types.GetPromptResult:
if name not in PROMPTS:
raise ValueError(f"Prompt not found: {name}")
if name == "git-commit":
changes = arguments.get("changes") if arguments else ""
return types.GetPromptResult(
messages=[
types.PromptMessage(
role="user",
content=types.TextContent(
type="text",
text=f"Generate a concise but descriptive commit message "
f"for these changes:\n\n{changes}"
)
)
]
)
if name == "explain-code":
code = arguments.get("code") if arguments else ""
language = arguments.get("language", "Unknown") if arguments else "Unknown"
return types.GetPromptResult(
messages=[
types.PromptMessage(
role="user",
content=types.TextContent(
type="text",
text=f"Explain how this {language} code works:\n\n{code}"
)
)
]
)
raise ValueError("Prompt implementation not found")
```
</Tab>
</Tabs>
## Best practices
When implementing prompts:
1. Use clear, descriptive prompt names
2. Provide detailed descriptions for prompts and arguments
3. Validate all required arguments
4. Handle missing arguments gracefully
5. Consider versioning for prompt templates
6. Cache dynamic content when appropriate
7. Implement error handling
8. Document expected argument formats
9. Consider prompt composability
10. Test prompts with various inputs
## UI integration
Prompts can be surfaced in client UIs as:
* Slash commands
* Quick actions
* Context menu items
* Command palette entries
* Guided workflows
* Interactive forms
## Updates and changes
Servers can notify clients about prompt changes:
1. Server capability: `prompts.listChanged`
2. Notification: `notifications/prompts/list_changed`
3. Client re-fetches prompt list
## Security considerations
When implementing prompts:
* Validate all arguments
* Sanitize user input
* Consider rate limiting
* Implement access controls
* Audit prompt usage
* Handle sensitive data appropriately
* Validate generated content
* Implement timeouts
* Consider prompt injection risks
* Document security requirements
# Resources
Expose data and content from your servers to LLMs
Resources are a core primitive in the Model Context Protocol (MCP) that allow servers to expose data and content that can be read by clients and used as context for LLM interactions.
<Note>
Resources are designed to be **application-controlled**, meaning that the client application can decide how and when they should be used.
Different MCP clients may handle resources differently. For example:
* Claude Desktop currently requires users to explicitly select resources before they can be used
* Other clients might automatically select resources based on heuristics
* Some implementations may even allow the AI model itself to determine which resources to use
Server authors should be prepared to handle any of these interaction patterns when implementing resource support. In order to expose data to models automatically, server authors should use a **model-controlled** primitive such as [Tools](./tools).
</Note>
## Overview
Resources represent any kind of data that an MCP server wants to make available to clients. This can include:
* File contents
* Database records
* API responses
* Live system data
* Screenshots and images
* Log files
* And more
Each resource is identified by a unique URI and can contain either text or binary data.
## Resource URIs
Resources are identified using URIs that follow this format:
```
[protocol]://[host]/[path]
```
For example:
* `file:///home/user/documents/report.pdf`
* `postgres://database/customers/schema`
* `screen://localhost/display1`
The protocol and path structure is defined by the MCP server implementation. Servers can define their own custom URI schemes.
## Resource types
Resources can contain two types of content:
### Text resources
Text resources contain UTF-8 encoded text data. These are suitable for:
* Source code
* Configuration files
* Log files
* JSON/XML data
* Plain text
### Binary resources
Binary resources contain raw binary data encoded in base64. These are suitable for:
* Images
* PDFs
* Audio files
* Video files
* Other non-text formats
## Resource discovery
Clients can discover available resources through two main methods:
### Direct resources
Servers expose a list of concrete resources via the `resources/list` endpoint. Each resource includes:
```typescript
{
uri: string; // Unique identifier for the resource
name: string; // Human-readable name
description?: string; // Optional description
mimeType?: string; // Optional MIME type
}
```
### Resource templates
For dynamic resources, servers can expose [URI templates](https://datatracker.ietf.org/doc/html/rfc6570) that clients can use to construct valid resource URIs:
```typescript
{
uriTemplate: string; // URI template following RFC 6570
name: string; // Human-readable name for this type
description?: string; // Optional description
mimeType?: string; // Optional MIME type for all matching resources
}
```
## Reading resources
To read a resource, clients make a `resources/read` request with the resource URI.
The server responds with a list of resource contents:
```typescript
{
contents: [
{
uri: string; // The URI of the resource
mimeType?: string; // Optional MIME type
// One of:
text?: string; // For text resources
blob?: string; // For binary resources (base64 encoded)
}
]
}
```
<Tip>
Servers may return multiple resources in response to one `resources/read` request. This could be used, for example, to return a list of files inside a directory when the directory is read.
</Tip>
## Resource updates
MCP supports real-time updates for resources through two mechanisms:
### List changes
Servers can notify clients when their list of available resources changes via the `notifications/resources/list_changed` notification.
### Content changes
Clients can subscribe to updates for specific resources:
1. Client sends `resources/subscribe` with resource URI
2. Server sends `notifications/resources/updated` when the resource changes
3. Client can fetch latest content with `resources/read`
4. Client can unsubscribe with `resources/unsubscribe`
## Example implementation
Here's a simple example of implementing resource support in an MCP server:
<Tabs>
<Tab title="TypeScript">
```typescript
const server = new Server({
name: "example-server",
version: "1.0.0"
}, {
capabilities: {
resources: {}
}
});
// List available resources
server.setRequestHandler(ListResourcesRequestSchema, async () => {
return {
resources: [
{
uri: "file:///logs/app.log",
name: "Application Logs",
mimeType: "text/plain"
}
]
};
});
// Read resource contents
server.setRequestHandler(ReadResourceRequestSchema, async (request) => {
const uri = request.params.uri;
if (uri === "file:///logs/app.log") {
const logContents = await readLogFile();
return {
contents: [
{
uri,
mimeType: "text/plain",
text: logContents
}
]
};
}
throw new Error("Resource not found");
});
```
</Tab>
<Tab title="Python">
```python
app = Server("example-server")
@app.list_resources()
async def list_resources() -> list[types.Resource]:
return [
types.Resource(
uri="file:///logs/app.log",
name="Application Logs",
mimeType="text/plain"
)
]
@app.read_resource()
async def read_resource(uri: AnyUrl) -> str:
if str(uri) == "file:///logs/app.log":
log_contents = await read_log_file()
return log_contents
raise ValueError("Resource not found")
# Start server
async with stdio_server() as streams:
await app.run(
streams[0],
streams[1],
app.create_initialization_options()
)
```
</Tab>
</Tabs>
## Best practices
When implementing resource support:
1. Use clear, descriptive resource names and URIs
2. Include helpful descriptions to guide LLM understanding
3. Set appropriate MIME types when known
4. Implement resource templates for dynamic content
5. Use subscriptions for frequently changing resources
6. Handle errors gracefully with clear error messages
7. Consider pagination for large resource lists
8. Cache resource contents when appropriate
9. Validate URIs before processing
10. Document your custom URI schemes
## Security considerations
When exposing resources:
* Validate all resource URIs
* Implement appropriate access controls
* Sanitize file paths to prevent directory traversal
* Be cautious with binary data handling
* Consider rate limiting for resource reads
* Audit resource access
* Encrypt sensitive data in transit
* Validate MIME types
* Implement timeouts for long-running reads
* Handle resource cleanup appropriately
# Sampling
Let your servers request completions from LLMs
Sampling is a powerful MCP feature that allows servers to request LLM completions through the client, enabling sophisticated agentic behaviors while maintaining security and privacy.
<Info>
This feature of MCP is not yet supported in the Claude Desktop client.
</Info>
## How sampling works
The sampling flow follows these steps:
1. Server sends a `sampling/createMessage` request to the client
2. Client reviews the request and can modify it
3. Client samples from an LLM
4. Client reviews the completion
5. Client returns the result to the server
This human-in-the-loop design ensures users maintain control over what the LLM sees and generates.
## Message format
Sampling requests use a standardized message format:
```typescript
{
messages: [
{
role: "user" | "assistant",
content: {
type: "text" | "image",
// For text:
text?: string,
// For images:
data?: string, // base64 encoded
mimeType?: string
}
}
],
modelPreferences?: {
hints?: [{
name?: string // Suggested model name/family
}],
costPriority?: number, // 0-1, importance of minimizing cost
speedPriority?: number, // 0-1, importance of low latency
intelligencePriority?: number // 0-1, importance of capabilities
},
systemPrompt?: string,
includeContext?: "none" | "thisServer" | "allServers",
temperature?: number,
maxTokens: number,
stopSequences?: string[],
metadata?: Record<string, unknown>
}
```
## Request parameters
### Messages
The `messages` array contains the conversation history to send to the LLM. Each message has:
* `role`: Either "user" or "assistant"
* `content`: The message content, which can be:
* Text content with a `text` field
* Image content with `data` (base64) and `mimeType` fields
### Model preferences
The `modelPreferences` object allows servers to specify their model selection preferences:
* `hints`: Array of model name suggestions that clients can use to select an appropriate model:
* `name`: String that can match full or partial model names (e.g. "claude-3", "sonnet")
* Clients may map hints to equivalent models from different providers
* Multiple hints are evaluated in preference order
* Priority values (0-1 normalized):
* `costPriority`: Importance of minimizing costs
* `speedPriority`: Importance of low latency response
* `intelligencePriority`: Importance of advanced model capabilities
Clients make the final model selection based on these preferences and their available models.
### System prompt
An optional `systemPrompt` field allows servers to request a specific system prompt. The client may modify or ignore this.
### Context inclusion
The `includeContext` parameter specifies what MCP context to include:
* `"none"`: No additional context
* `"thisServer"`: Include context from the requesting server
* `"allServers"`: Include context from all connected MCP servers
The client controls what context is actually included.
### Sampling parameters
Fine-tune the LLM sampling with:
* `temperature`: Controls randomness (0.0 to 1.0)
* `maxTokens`: Maximum tokens to generate
* `stopSequences`: Array of sequences that stop generation
* `metadata`: Additional provider-specific parameters
## Response format
The client returns a completion result:
```typescript
{
model: string, // Name of the model used
stopReason?: "endTurn" | "stopSequence" | "maxTokens" | string,
role: "user" | "assistant",
content: {
type: "text" | "image",
text?: string,
data?: string,
mimeType?: string
}
}
```
## Example request
Here's an example of requesting sampling from a client:
```json
{
"method": "sampling/createMessage",
"params": {
"messages": [
{
"role": "user",
"content": {
"type": "text",
"text": "What files are in the current directory?"
}
}
],
"systemPrompt": "You are a helpful file system assistant.",
"includeContext": "thisServer",
"maxTokens": 100
}
}
```
## Best practices
When implementing sampling:
1. Always provide clear, well-structured prompts
2. Handle both text and image content appropriately
3. Set reasonable token limits
4. Include relevant context through `includeContext`
5. Validate responses before using them
6. Handle errors gracefully
7. Consider rate limiting sampling requests
8. Document expected sampling behavior
9. Test with various model parameters
10. Monitor sampling costs
## Human in the loop controls
Sampling is designed with human oversight in mind:
### For prompts
* Clients should show users the proposed prompt
* Users should be able to modify or reject prompts
* System prompts can be filtered or modified
* Context inclusion is controlled by the client
### For completions
* Clients should show users the completion
* Users should be able to modify or reject completions
* Clients can filter or modify completions
* Users control which model is used
## Security considerations
When implementing sampling:
* Validate all message content
* Sanitize sensitive information
* Implement appropriate rate limits
* Monitor sampling usage
* Encrypt data in transit
* Handle user data privacy
* Audit sampling requests
* Control cost exposure
* Implement timeouts
* Handle model errors gracefully
## Common patterns
### Agentic workflows
Sampling enables agentic patterns like:
* Reading and analyzing resources
* Making decisions based on context
* Generating structured data
* Handling multi-step tasks
* Providing interactive assistance
### Context management
Best practices for context:
* Request minimal necessary context
* Structure context clearly
* Handle context size limits
* Update context as needed
* Clean up stale context
### Error handling
Robust error handling should:
* Catch sampling failures
* Handle timeout errors
* Manage rate limits
* Validate responses
* Provide fallback behaviors
* Log errors appropriately
## Limitations
Be aware of these limitations:
* Sampling depends on client capabilities
* Users control sampling behavior
* Context size has limits
* Rate limits may apply
* Costs should be considered
* Model availability varies
* Response times vary
* Not all content types supported
# Tools
Enable LLMs to perform actions through your server
Tools are a powerful primitive in the Model Context Protocol (MCP) that enable servers to expose executable functionality to clients. Through tools, LLMs can interact with external systems, perform computations, and take actions in the real world.
<Note>
Tools are designed to be **model-controlled**, meaning that tools are exposed from servers to clients with the intention of the AI model being able to automatically invoke them (with a human in the loop to grant approval).
</Note>
## Overview
Tools in MCP allow servers to expose executable functions that can be invoked by clients and used by LLMs to perform actions. Key aspects of tools include:
* **Discovery**: Clients can list available tools through the `tools/list` endpoint
* **Invocation**: Tools are called using the `tools/call` endpoint, where servers perform the requested operation and return results
* **Flexibility**: Tools can range from simple calculations to complex API interactions
Like [resources](/docs/concepts/resources), tools are identified by unique names and can include descriptions to guide their usage. However, unlike resources, tools represent dynamic operations that can modify state or interact with external systems.
## Tool definition structure
Each tool is defined with the following structure:
```typescript
{
name: string; // Unique identifier for the tool
description?: string; // Human-readable description
inputSchema: { // JSON Schema for the tool's parameters
type: "object",
properties: { ... } // Tool-specific parameters
}
}
```
## Implementing tools
Here's an example of implementing a basic tool in an MCP server:
<Tabs>
<Tab title="TypeScript">
```typescript
const server = new Server({
name: "example-server",
version: "1.0.0"
}, {
capabilities: {
tools: {}
}
});
// Define available tools
server.setRequestHandler(ListToolsRequestSchema, async () => {
return {
tools: [{
name: "calculate_sum",
description: "Add two numbers together",
inputSchema: {
type: "object",
properties: {
a: { type: "number" },
b: { type: "number" }
},
required: ["a", "b"]
}
}]
};
});
// Handle tool execution
server.setRequestHandler(CallToolRequestSchema, async (request) => {
if (request.params.name === "calculate_sum") {
const { a, b } = request.params.arguments;
return {
toolResult: a + b
};
}
throw new Error("Tool not found");
});
```
</Tab>
<Tab title="Python">
```python
app = Server("example-server")
@app.list_tools()
async def list_tools() -> list[types.Tool]:
return [
types.Tool(
name="calculate_sum",
description="Add two numbers together",
inputSchema={
"type": "object",
"properties": {
"a": {"type": "number"},
"b": {"type": "number"}
},
"required": ["a", "b"]
}
)
]
@app.call_tool()
async def call_tool(
name: str,
arguments: dict
) -> list[types.TextContent | types.ImageContent | types.EmbeddedResource]:
if name == "calculate_sum":
a = arguments["a"]
b = arguments["b"]
result = a + b
return [types.TextContent(type="text", text=str(result))]
raise ValueError(f"Tool not found: {name}")
```
</Tab>
</Tabs>
## Example tool patterns
Here are some examples of types of tools that a server could provide:
### System operations
Tools that interact with the local system:
```typescript
{
name: "execute_command",
description: "Run a shell command",
inputSchema: {
type: "object",
properties: {
command: { type: "string" },
args: { type: "array", items: { type: "string" } }
}
}
}
```
### API integrations
Tools that wrap external APIs:
```typescript
{
name: "github_create_issue",
description: "Create a GitHub issue",
inputSchema: {
type: "object",
properties: {
title: { type: "string" },
body: { type: "string" },
labels: { type: "array", items: { type: "string" } }
}
}
}
```
### Data processing
Tools that transform or analyze data:
```typescript
{
name: "analyze_csv",
description: "Analyze a CSV file",
inputSchema: {
type: "object",
properties: {
filepath: { type: "string" },
operations: {
type: "array",
items: {
enum: ["sum", "average", "count"]
}
}
}
}
}
```
## Best practices
When implementing tools:
1. Provide clear, descriptive names and descriptions
2. Use detailed JSON Schema definitions for parameters
3. Include examples in tool descriptions to demonstrate how the model should use them
4. Implement proper error handling and validation
5. Use progress reporting for long operations
6. Keep tool operations focused and atomic
7. Document expected return value structures
8. Implement proper timeouts
9. Consider rate limiting for resource-intensive operations
10. Log tool usage for debugging and monitoring
## Security considerations
When exposing tools:
### Input validation
* Validate all parameters against the schema
* Sanitize file paths and system commands
* Validate URLs and external identifiers
* Check parameter sizes and ranges
* Prevent command injection
### Access control
* Implement authentication where needed
* Use appropriate authorization checks
* Audit tool usage
* Rate limit requests
* Monitor for abuse
### Error handling
* Don't expose internal errors to clients
* Log security-relevant errors
* Handle timeouts appropriately
* Clean up resources after errors
* Validate return values
## Tool discovery and updates
MCP supports dynamic tool discovery:
1. Clients can list available tools at any time
2. Servers can notify clients when tools change using `notifications/tools/list_changed`
3. Tools can be added or removed during runtime
4. Tool definitions can be updated (though this should be done carefully)
## Error handling
Tool errors should be reported within the result object, not as MCP protocol-level errors. This allows the LLM to see and potentially handle the error. When a tool encounters an error:
1. Set `isError` to `true` in the result
2. Include error details in the `content` array
Here's an example of proper error handling for tools:
<Tabs>
<Tab title="TypeScript">
```typescript
try {
// Tool operation
const result = performOperation();
return {
content: [
{
type: "text",
text: `Operation successful: ${result}`
}
]
};
} catch (error) {
return {
isError: true,
content: [
{
type: "text",
text: `Error: ${error.message}`
}
]
};
}
```
</Tab>
<Tab title="Python">
```python
try:
# Tool operation
result = perform_operation()
return types.CallToolResult(
content=[
types.TextContent(
type="text",
text=f"Operation successful: {result}"
)
]
)
except Exception as error:
return types.CallToolResult(
isError=True,
content=[
types.TextContent(
type="text",
text=f"Error: {str(error)}"
)
]
)
```
</Tab>
</Tabs>
This approach allows the LLM to see that an error occurred and potentially take corrective action or request human intervention.
## Testing tools
A comprehensive testing strategy for MCP tools should cover:
* **Functional testing**: Verify tools execute correctly with valid inputs and handle invalid inputs appropriately
* **Integration testing**: Test tool interaction with external systems using both real and mocked dependencies
* **Security testing**: Validate authentication, authorization, input sanitization, and rate limiting
* **Performance testing**: Check behavior under load, timeout handling, and resource cleanup
* **Error handling**: Ensure tools properly report errors through the MCP protocol and clean up resources
# Transports
Learn about MCP's communication mechanisms
Transports in the Model Context Protocol (MCP) provide the foundation for communication between clients and servers. A transport handles the underlying mechanics of how messages are sent and received.
## Message Format
MCP uses [JSON-RPC](https://www.jsonrpc.org/) 2.0 as its wire format. The transport layer is responsible for converting MCP protocol messages into JSON-RPC format for transmission and converting received JSON-RPC messages back into MCP protocol messages.
There are three types of JSON-RPC messages used:
### Requests
```typescript
{
jsonrpc: "2.0",
id: number | string,
method: string,
params?: object
}
```
### Responses
```typescript
{
jsonrpc: "2.0",
id: number | string,
result?: object,
error?: {
code: number,
message: string,
data?: unknown
}
}
```
### Notifications
```typescript
{
jsonrpc: "2.0",
method: string,
params?: object
}
```
## Built-in Transport Types
MCP includes two standard transport implementations:
### Standard Input/Output (stdio)
The stdio transport enables communication through standard input and output streams. This is particularly useful for local integrations and command-line tools.
Use stdio when:
* Building command-line tools
* Implementing local integrations
* Needing simple process communication
* Working with shell scripts
<Tabs>
<Tab title="TypeScript (Server)">
```typescript
const server = new Server({
name: "example-server",
version: "1.0.0"
}, {
capabilities: {}
});
const transport = new StdioServerTransport();
await server.connect(transport);
```
</Tab>
<Tab title="TypeScript (Client)">
```typescript
const client = new Client({
name: "example-client",
version: "1.0.0"
}, {
capabilities: {}
});
const transport = new StdioClientTransport({
command: "./server",
args: ["--option", "value"]
});
await client.connect(transport);
```
</Tab>
<Tab title="Python (Server)">
```python
app = Server("example-server")
async with stdio_server() as streams:
await app.run(
streams[0],
streams[1],
app.create_initialization_options()
)
```
</Tab>
<Tab title="Python (Client)">
```python
params = StdioServerParameters(
command="./server",
args=["--option", "value"]
)
async with stdio_client(params) as streams:
async with ClientSession(streams[0], streams[1]) as session:
await session.initialize()
```
</Tab>
</Tabs>
### Server-Sent Events (SSE)
SSE transport enables server-to-client streaming with HTTP POST requests for client-to-server communication.
Use SSE when:
* Only server-to-client streaming is needed
* Working with restricted networks
* Implementing simple updates
<Tabs>
<Tab title="TypeScript (Server)">
```typescript
const server = new Server({
name: "example-server",
version: "1.0.0"
}, {
capabilities: {}
});
const transport = new SSEServerTransport("/message", response);
await server.connect(transport);
```
</Tab>
<Tab title="TypeScript (Client)">
```typescript
const client = new Client({
name: "example-client",
version: "1.0.0"
}, {
capabilities: {}
});
const transport = new SSEClientTransport(
new URL("http://localhost:3000/sse")
);
await client.connect(transport);
```
</Tab>
<Tab title="Python (Server)">
```python
from mcp.server.sse import SseServerTransport
from starlette.applications import Starlette
from starlette.routing import Route
app = Server("example-server")
sse = SseServerTransport("/messages")
async def handle_sse(scope, receive, send):
async with sse.connect_sse(scope, receive, send) as streams:
await app.run(streams[0], streams[1], app.create_initialization_options())
async def handle_messages(scope, receive, send):
await sse.handle_post_message(scope, receive, send)
starlette_app = Starlette(
routes=[
Route("/sse", endpoint=handle_sse),
Route("/messages", endpoint=handle_messages, methods=["POST"]),
]
)
```
</Tab>
<Tab title="Python (Client)">
```python
async with sse_client("http://localhost:8000/sse") as streams:
async with ClientSession(streams[0], streams[1]) as session:
await session.initialize()
```
</Tab>
</Tabs>
## Custom Transports
MCP makes it easy to implement custom transports for specific needs. Any transport implementation just needs to conform to the Transport interface:
You can implement custom transports for:
* Custom network protocols
* Specialized communication channels
* Integration with existing systems
* Performance optimization
<Tabs>
<Tab title="TypeScript">
```typescript
interface Transport {
// Start processing messages
start(): Promise<void>;
// Send a JSON-RPC message
send(message: JSONRPCMessage): Promise<void>;
// Close the connection
close(): Promise<void>;
// Callbacks
onclose?: () => void;
onerror?: (error: Error) => void;
onmessage?: (message: JSONRPCMessage) => void;
}
```
</Tab>
<Tab title="Python">
Note that while MCP Servers are often implemented with asyncio, we recommend
implementing low-level interfaces like transports with `anyio` for wider compatibility.
```python
@contextmanager
async def create_transport(
read_stream: MemoryObjectReceiveStream[JSONRPCMessage | Exception],
write_stream: MemoryObjectSendStream[JSONRPCMessage]
):
"""
Transport interface for MCP.
Args:
read_stream: Stream to read incoming messages from
write_stream: Stream to write outgoing messages to
"""
async with anyio.create_task_group() as tg:
try:
# Start processing messages
tg.start_soon(lambda: process_messages(read_stream))
# Send messages
async with write_stream:
yield write_stream
except Exception as exc:
# Handle errors
raise exc
finally:
# Clean up
tg.cancel_scope.cancel()
await write_stream.aclose()
await read_stream.aclose()
```
</Tab>
</Tabs>
## Error Handling
Transport implementations should handle various error scenarios:
1. Connection errors
2. Message parsing errors
3. Protocol errors
4. Network timeouts
5. Resource cleanup
Example error handling:
<Tabs>
<Tab title="TypeScript">
```typescript
class ExampleTransport implements Transport {
async start() {
try {
// Connection logic
} catch (error) {
this.onerror?.(new Error(`Failed to connect: ${error}`));
throw error;
}
}
async send(message: JSONRPCMessage) {
try {
// Sending logic
} catch (error) {
this.onerror?.(new Error(`Failed to send message: ${error}`));
throw error;
}
}
}
```
</Tab>
<Tab title="Python">
Note that while MCP Servers are often implemented with asyncio, we recommend
implementing low-level interfaces like transports with `anyio` for wider compatibility.
```python
@contextmanager
async def example_transport(scope: Scope, receive: Receive, send: Send):
try:
# Create streams for bidirectional communication
read_stream_writer, read_stream = anyio.create_memory_object_stream(0)
write_stream, write_stream_reader = anyio.create_memory_object_stream(0)
async def message_handler():
try:
async with read_stream_writer:
# Message handling logic
pass
except Exception as exc:
logger.error(f"Failed to handle message: {exc}")
raise exc
async with anyio.create_task_group() as tg:
tg.start_soon(message_handler)
try:
# Yield streams for communication
yield read_stream, write_stream
except Exception as exc:
logger.error(f"Transport error: {exc}")
raise exc
finally:
tg.cancel_scope.cancel()
await write_stream.aclose()
await read_stream.aclose()
except Exception as exc:
logger.error(f"Failed to initialize transport: {exc}")
raise exc
```
</Tab>
</Tabs>
## Best Practices
When implementing or using MCP transport:
1. Handle connection lifecycle properly
2. Implement proper error handling
3. Clean up resources on connection close
4. Use appropriate timeouts
5. Validate messages before sending
6. Log transport events for debugging
7. Implement reconnection logic when appropriate
8. Handle backpressure in message queues
9. Monitor connection health
10. Implement proper security measures
## Security Considerations
When implementing transport:
### Authentication and Authorization
* Implement proper authentication mechanisms
* Validate client credentials
* Use secure token handling
* Implement authorization checks
### Data Security
* Use TLS for network transport
* Encrypt sensitive data
* Validate message integrity
* Implement message size limits
* Sanitize input data
### Network Security
* Implement rate limiting
* Use appropriate timeouts
* Handle denial of service scenarios
* Monitor for unusual patterns
* Implement proper firewall rules
## Debugging Transport
Tips for debugging transport issues:
1. Enable debug logging
2. Monitor message flow
3. Check connection states
4. Validate message formats
5. Test error scenarios
6. Use network analysis tools
7. Implement health checks
8. Monitor resource usage
9. Test edge cases
10. Use proper error tracking
# Python
Create a simple MCP server in Python in 15 minutes
Let's build your first MCP server in Python! We'll create a weather server that provides current weather data as a resource and lets Claude fetch forecasts using tools.
<Note>
This guide uses the OpenWeatherMap API. You'll need a free API key from [OpenWeatherMap](https://openweathermap.org/api) to follow along.
</Note>
## Prerequisites
<Info>
The following steps are for macOS. Guides for other platforms are coming soon.
</Info>
<Steps>
<Step title="Install Python">
You'll need Python 3.10 or higher:
```bash
python --version # Should be 3.10 or higher
```
</Step>
<Step title="Install uv via homebrew">
See [https://docs.astral.sh/uv/](https://docs.astral.sh/uv/) for more information.
```bash
brew install uv
uv --version # Should be 0.4.18 or higher
```
</Step>
<Step title="Create a new project using the MCP project creator">
```bash
uvx create-mcp-server --path weather_service
cd weather_service
```
</Step>
<Step title="Install additional dependencies">
```bash
uv add httpx python-dotenv
```
</Step>
<Step title="Set up environment">
Create `.env`:
```bash
OPENWEATHER_API_KEY=your-api-key-here
```
</Step>
</Steps>
## Create your server
<Steps>
<Step title="Add the base imports and setup">
In `weather_service/src/weather_service/server.py`
```python
import os
import json
import logging
from datetime import datetime, timedelta
from collections.abc import Sequence
from functools import lru_cache
from typing import Any
import httpx
import asyncio
from dotenv import load_dotenv
from mcp.server import Server
from mcp.types import (
Resource,
Tool,
TextContent,
ImageContent,
EmbeddedResource,
LoggingLevel
)
from pydantic import AnyUrl
# Load environment variables
load_dotenv()
# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger("weather-server")
# API configuration
API_KEY = os.getenv("OPENWEATHER_API_KEY")
if not API_KEY:
raise ValueError("OPENWEATHER_API_KEY environment variable required")
API_BASE_URL = "http://api.openweathermap.org/data/2.5"
DEFAULT_CITY = "London"
CURRENT_WEATHER_ENDPOINT = "weather"
FORECAST_ENDPOINT = "forecast"
# The rest of our server implementation will go here
```
</Step>
<Step title="Add weather fetching functionality">
Add this functionality:
```python
# Create reusable params
http_params = {
"appid": API_KEY,
"units": "metric"
}
async def fetch_weather(city: str) -> dict[str, Any]:
async with httpx.AsyncClient() as client:
response = await client.get(
f"{API_BASE_URL}/weather",
params={"q": city, **http_params}
)
response.raise_for_status()
data = response.json()
return {
"temperature": data["main"]["temp"],
"conditions": data["weather"][0]["description"],
"humidity": data["main"]["humidity"],
"wind_speed": data["wind"]["speed"],
"timestamp": datetime.now().isoformat()
}
app = Server("weather-server")
```
</Step>
<Step title="Implement resource handlers">
Add these resource-related handlers to our main function:
```python
app = Server("weather-server")
@app.list_resources()
async def list_resources() -> list[Resource]:
"""List available weather resources."""
uri = AnyUrl(f"weather://{DEFAULT_CITY}/current")
return [
Resource(
uri=uri,
name=f"Current weather in {DEFAULT_CITY}",
mimeType="application/json",
description="Real-time weather data"
)
]
@app.read_resource()
async def read_resource(uri: AnyUrl) -> str:
"""Read current weather data for a city."""
city = DEFAULT_CITY
if str(uri).startswith("weather://") and str(uri).endswith("/current"):
city = str(uri).split("/")[-2]
else:
raise ValueError(f"Unknown resource: {uri}")
try:
weather_data = await fetch_weather(city)
return json.dumps(weather_data, indent=2)
except httpx.HTTPError as e:
raise RuntimeError(f"Weather API error: {str(e)}")
```
</Step>
<Step title="Implement tool handlers">
Add these tool-related handlers:
```python
app = Server("weather-server")
# Resource implementation ...
@app.list_tools()
async def list_tools() -> list[Tool]:
"""List available weather tools."""
return [
Tool(
name="get_forecast",
description="Get weather forecast for a city",
inputSchema={
"type": "object",
"properties": {
"city": {
"type": "string",
"description": "City name"
},
"days": {
"type": "number",
"description": "Number of days (1-5)",
"minimum": 1,
"maximum": 5
}
},
"required": ["city"]
}
)
]
@app.call_tool()
async def call_tool(name: str, arguments: Any) -> Sequence[TextContent | ImageContent | EmbeddedResource]:
"""Handle tool calls for weather forecasts."""
if name != "get_forecast":
raise ValueError(f"Unknown tool: {name}")
if not isinstance(arguments, dict) or "city" not in arguments:
raise ValueError("Invalid forecast arguments")
city = arguments["city"]
days = min(int(arguments.get("days", 3)), 5)
try:
async with httpx.AsyncClient() as client:
response = await client.get(
f"{API_BASE_URL}/{FORECAST_ENDPOINT}",
params={
"q": city,
"cnt": days * 8, # API returns 3-hour intervals
**http_params,
}
)
response.raise_for_status()
data = response.json()
forecasts = []
for i in range(0, len(data["list"]), 8):
day_data = data["list"][i]
forecasts.append({
"date": day_data["dt_txt"].split()[0],
"temperature": day_data["main"]["temp"],
"conditions": day_data["weather"][0]["description"]
})
return [
TextContent(
type="text",
text=json.dumps(forecasts, indent=2)
)
]
except httpx.HTTPError as e:
logger.error(f"Weather API error: {str(e)}")
raise RuntimeError(f"Weather API error: {str(e)}")
```
</Step>
<Step title="Add the main function">
Add this to the end of `weather_service/src/weather_service/server.py`:
```python
async def main():
# Import here to avoid issues with event loops
from mcp.server.stdio import stdio_server
async with stdio_server() as (read_stream, write_stream):
await app.run(
read_stream,
write_stream,
app.create_initialization_options()
)
```
</Step>
<Step title="Check your entry point in __init__.py">
Add this to the end of `weather_service/src/weather_service/__init__.py`:
```python
from . import server
import asyncio
def main():
"""Main entry point for the package."""
asyncio.run(server.main())
# Optionally expose other important items at package level
__all__ = ['main', 'server']
```
</Step>
</Steps>
## Connect to Claude Desktop
<Steps>
<Step title="Update Claude config">
Add to `claude_desktop_config.json`:
```json
{
"mcpServers": {
"weather": {
"command": "uv",
"args": [
"--directory",
"path/to/your/project",
"run",
"weather-service"
],
"env": {
"OPENWEATHER_API_KEY": "your-api-key"
}
}
}
}
```
</Step>
<Step title="Restart Claude">
1. Quit Claude completely
2. Start Claude again
3. Look for your weather server in the 🔌 menu
</Step>
</Steps>
## Try it out!
<AccordionGroup>
<Accordion title="Check Current Weather" active>
Ask Claude:
```
What's the current weather in San Francisco? Can you analyze the conditions and tell me if it's a good day for outdoor activities?
```
</Accordion>
<Accordion title="Get a Forecast">
Ask Claude:
```
Can you get me a 5-day forecast for Tokyo and help me plan what clothes to pack for my trip?
```
</Accordion>
<Accordion title="Compare Weather">
Ask Claude:
```
Can you analyze the forecast for both Tokyo and San Francisco and tell me which city would be better for outdoor photography this week?
```
</Accordion>
</AccordionGroup>
## Understanding the code
<Tabs>
<Tab title="Type Hints">
```python
async def read_resource(self, uri: str) -> ReadResourceResult:
# ...
```
Python type hints help catch errors early and improve code maintainability.
</Tab>
<Tab title="Resources">
```python
@app.list_resources()
async def list_resources(self) -> ListResourcesResult:
return ListResourcesResult(
resources=[
Resource(
uri=f"weather://{DEFAULT_CITY}/current",
name=f"Current weather in {DEFAULT_CITY}",
mimeType="application/json",
description="Real-time weather data"
)
]
)
```
Resources provide data that Claude can access as context.
</Tab>
<Tab title="Tools">
```python
Tool(
name="get_forecast",
description="Get weather forecast for a city",
inputSchema={
"type": "object",
"properties": {
"city": {
"type": "string",
"description": "City name"
},
"days": {
"type": "number",
"description": "Number of days (1-5)",
"minimum": 1,
"maximum": 5
}
},
"required": ["city"]
}
)
```
Tools let Claude take actions through your server with validated inputs.
</Tab>
<Tab title="Server Structure">
```python
# Create server instance with name
app = Server("weather-server")
# Register resource handler
@app.list_resources()
async def list_resources() -> list[Resource]:
"""List available resources"""
return [...]
# Register tool handler
@app.call_tool()
async def call_tool(name: str, arguments: Any) -> Sequence[TextContent]:
"""Handle tool execution"""
return [...]
# Register additional handlers
@app.read_resource()
...
@app.list_tools()
...
```
The MCP server uses a simple app pattern - create a Server instance and register handlers with decorators. Each handler maps to a specific MCP protocol operation.
</Tab>
</Tabs>
## Best practices
<CardGroup cols={1}>
<Card title="Error Handling" icon="shield">
```python
try:
async with httpx.AsyncClient() as client:
response = await client.get(..., params={..., **http_params})
response.raise_for_status()
except httpx.HTTPError as e:
raise McpError(
ErrorCode.INTERNAL_ERROR,
f"API error: {str(e)}"
)
```
</Card>
<Card title="Type Validation" icon="check">
```python
if not isinstance(args, dict) or "city" not in args:
raise McpError(
ErrorCode.INVALID_PARAMS,
"Invalid forecast arguments"
)
```
</Card>
<Card title="Environment Variables" icon="gear">
```python
if not API_KEY:
raise ValueError("OPENWEATHER_API_KEY is required")
```
</Card>
</CardGroup>
## Available transports
While this guide uses stdio transport, MCP supports additional transport options:
### SSE (Server-Sent Events)
```python
from mcp.server.sse import SseServerTransport
from starlette.applications import Starlette
from starlette.routing import Route
# Create SSE transport with endpoint
sse = SseServerTransport("/messages")
# Handler for SSE connections
async def handle_sse(scope, receive, send):
async with sse.connect_sse(scope, receive, send) as streams:
await app.run(
streams[0], streams[1], app.create_initialization_options()
)
# Handler for client messages
async def handle_messages(scope, receive, send):
await sse.handle_post_message(scope, receive, send)
# Create Starlette app with routes
app = Starlette(
debug=True,
routes=[
Route("/sse", endpoint=handle_sse),
Route("/messages", endpoint=handle_messages, methods=["POST"]),
],
)
# Run with any ASGI server
import uvicorn
uvicorn.run(app, host="0.0.0.0", port=8000)
```
## Advanced features
<Steps>
<Step title="Understanding Request Context">
The request context provides access to the current request's metadata and the active client session. Access it through `server.request_context`:
```python
@app.call_tool()
async def call_tool(name: str, arguments: Any) -> Sequence[TextContent]:
# Access the current request context
ctx = self.request_context
# Get request metadata like progress tokens
if progress_token := ctx.meta.progressToken:
# Send progress notifications via the session
await ctx.session.send_progress_notification(
progress_token=progress_token,
progress=0.5,
total=1.0
)
# Sample from the LLM client
result = await ctx.session.create_message(
messages=[
SamplingMessage(
role="user",
content=TextContent(
type="text",
text="Analyze this weather data: " + json.dumps(arguments)
)
)
],
max_tokens=100
)
return [TextContent(type="text", text=result.content.text)]
```
</Step>
<Step title="Add caching">
```python
# Cache settings
cache_timeout = timedelta(minutes=15)
last_cache_time = None
cached_weather = None
async def fetch_weather(city: str) -> dict[str, Any]:
global cached_weather, last_cache_time
now = datetime.now()
if (cached_weather is None or
last_cache_time is None or
now - last_cache_time > cache_timeout):
async with httpx.AsyncClient() as client:
response = await client.get(
f"{API_BASE_URL}/{CURRENT_WEATHER_ENDPOINT}",
params={"q": city, **http_params}
)
response.raise_for_status()
data = response.json()
cached_weather = {
"temperature": data["main"]["temp"],
"conditions": data["weather"][0]["description"],
"humidity": data["main"]["humidity"],
"wind_speed": data["wind"]["speed"],
"timestamp": datetime.now().isoformat()
}
last_cache_time = now
return cached_weather
```
</Step>
<Step title="Add progress notifications">
```python
@self.call_tool()
async def call_tool(self, name: str, arguments: Any) -> CallToolResult:
if progress_token := self.request_context.meta.progressToken:
# Send progress notifications
await self.request_context.session.send_progress_notification(
progress_token=progress_token,
progress=1,
total=2
)
# Fetch data...
await self.request_context.session.send_progress_notification(
progress_token=progress_token,
progress=2,
total=2
)
# Rest of the method implementation...
```
</Step>
<Step title="Add logging support">
```python
# Set up logging
logger = logging.getLogger("weather-server")
logger.setLevel(logging.INFO)
@app.set_logging_level()
async def set_logging_level(level: LoggingLevel) -> EmptyResult:
logger.setLevel(level.upper())
await app.request_context.session.send_log_message(
level="info",
data=f"Log level set to {level}",
logger="weather-server"
)
return EmptyResult()
# Use logger throughout the code
# For example:
# logger.info("Weather data fetched successfully")
# logger.error(f"Error fetching weather data: {str(e)}")
```
</Step>
<Step title="Add resource templates">
```python
@app.list_resource_templates()
async def list_resource_templates() -> list[ResourceTemplate]:
return [
ResourceTemplate(
uriTemplate="weather://{city}/current",
name="Current weather for any city",
mimeType="application/json"
)
]
```
</Step>
</Steps>
## Testing
<Steps>
<Step title="Create test file">
Create `tests/weather_test.py`:
```python
import pytest
import os
from unittest.mock import patch, Mock
from datetime import datetime
import json
from pydantic import AnyUrl
os.environ["OPENWEATHER_API_KEY"] = "TEST"
from weather_service.server import (
fetch_weather,
read_resource,
call_tool,
list_resources,
list_tools,
DEFAULT_CITY
)
@pytest.fixture
def anyio_backend():
return "asyncio"
@pytest.fixture
def mock_weather_response():
return {
"main": {
"temp": 20.5,
"humidity": 65
},
"weather": [
{"description": "scattered clouds"}
],
"wind": {
"speed": 3.6
}
}
@pytest.fixture
def mock_forecast_response():
return {
"list": [
{
"dt_txt": "2024-01-01 12:00:00",
"main": {"temp": 18.5},
"weather": [{"description": "sunny"}]
},
{
"dt_txt": "2024-01-02 12:00:00",
"main": {"temp": 17.2},
"weather": [{"description": "cloudy"}]
}
]
}
@pytest.mark.anyio
async def test_fetch_weather(mock_weather_response):
with patch('requests.Session.get') as mock_get:
mock_get.return_value.json.return_value = mock_weather_response
mock_get.return_value.raise_for_status = Mock()
weather = await fetch_weather("London")
assert weather["temperature"] == 20.5
assert weather["conditions"] == "scattered clouds"
assert weather["humidity"] == 65
assert weather["wind_speed"] == 3.6
assert "timestamp" in weather
@pytest.mark.anyio
async def test_read_resource():
with patch('weather_service.server.fetch_weather') as mock_fetch:
mock_fetch.return_value = {
"temperature": 20.5,
"conditions": "clear sky",
"timestamp": datetime.now().isoformat()
}
uri = AnyUrl("weather://London/current")
result = await read_resource(uri)
assert isinstance(result, str)
assert "temperature" in result
assert "clear sky" in result
@pytest.mark.anyio
async def test_call_tool(mock_forecast_response):
class Response():
def raise_for_status(self):
pass
def json(self):
return mock_forecast_response
class AsyncClient():
async def __aenter__(self):
return self
async def __aexit__(self, *exc_info):
pass
async def get(self, *args, **kwargs):
return Response()
with patch('httpx.AsyncClient', new=AsyncClient) as mock_client:
result = await call_tool("get_forecast", {"city": "London", "days": 2})
assert len(result) == 1
assert result[0].type == "text"
forecast_data = json.loads(result[0].text)
assert len(forecast_data) == 1
assert forecast_data[0]["temperature"] == 18.5
assert forecast_data[0]["conditions"] == "sunny"
@pytest.mark.anyio
async def test_list_resources():
resources = await list_resources()
assert len(resources) == 1
assert resources[0].name == f"Current weather in {DEFAULT_CITY}"
assert resources[0].mimeType == "application/json"
@pytest.mark.anyio
async def test_list_tools():
tools = await list_tools()
assert len(tools) == 1
assert tools[0].name == "get_forecast"
assert "city" in tools[0].inputSchema["properties"]
```
</Step>
<Step title="Run tests">
```bash
uv add --dev pytest
uv run pytest
```
</Step>
</Steps>
## Troubleshooting
### Installation issues
```bash
# Check Python version
python --version
# Reinstall dependencies
uv sync --reinstall
```
### Type checking
```bash
# Install mypy
uv add --dev pyright
# Run type checker
uv run pyright src
```
## Next steps
<CardGroup cols={2}>
<Card title="Architecture overview" icon="sitemap" href="/docs/concepts/architecture">
Learn more about the MCP architecture
</Card>
<Card title="Python SDK" icon="python" href="https://github.com/modelcontextprotocol/python-sdk">
Check out the Python SDK on GitHub
</Card>
</CardGroup>
# TypeScript
Create a simple MCP server in TypeScript in 15 minutes
Let's build your first MCP server in TypeScript! We'll create a weather server that provides current weather data as a resource and lets Claude fetch forecasts using tools.
<Note>
This guide uses the OpenWeatherMap API. You'll need a free API key from [OpenWeatherMap](https://openweathermap.org/api) to follow along.
</Note>
## Prerequisites
<Steps>
<Step title="Install Node.js">
You'll need Node.js 18 or higher:
```bash
node --version # Should be v18 or higher
npm --version
```
</Step>
<Step title="Create a new project">
You can use our [create-typescript-server](https://github.com/modelcontextprotocol/create-typescript-server) tool to bootstrap a new project:
```bash
npx @modelcontextprotocol/create-server weather-server
cd weather-server
```
</Step>
<Step title="Install dependencies">
```bash
npm install --save axios dotenv
```
</Step>
<Step title="Set up environment">
Create `.env`:
```bash
OPENWEATHER_API_KEY=your-api-key-here
```
Make sure to add your environment file to `.gitignore`
```bash
.env
```
</Step>
</Steps>
## Create your server
<Steps>
<Step title="Define types">
Create a file `src/types.ts`, and add the following:
```typescript
export interface OpenWeatherResponse {
main: {
temp: number;
humidity: number;
};
weather: Array<{
description: string;
}>;
wind: {
speed: number;
};
dt_txt?: string;
}
export interface WeatherData {
temperature: number;
conditions: string;
humidity: number;
wind_speed: number;
timestamp: string;
}
export interface ForecastDay {
date: string;
temperature: number;
conditions: string;
}
export interface GetForecastArgs {
city: string;
days?: number;
}
// Type guard for forecast arguments
export function isValidForecastArgs(args: any): args is GetForecastArgs {
return (
typeof args === "object" &&
args !== null &&
"city" in args &&
typeof args.city === "string" &&
(args.days === undefined || typeof args.days === "number")
);
}
```
</Step>
<Step title="Add the base code">
Replace `src/index.ts` with the following:
```typescript
#!/usr/bin/env node
import { Server } from "@modelcontextprotocol/sdk/server/index.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import {
ListResourcesRequestSchema,
ReadResourceRequestSchema,
ListToolsRequestSchema,
CallToolRequestSchema,
ErrorCode,
McpError
} from "@modelcontextprotocol/sdk/types.js";
import axios from "axios";
import dotenv from "dotenv";
import {
WeatherData,
ForecastDay,
OpenWeatherResponse,
isValidForecastArgs
} from "./types.js";
dotenv.config();
const API_KEY = process.env.OPENWEATHER_API_KEY;
if (!API_KEY) {
throw new Error("OPENWEATHER_API_KEY environment variable is required");
}
const API_CONFIG = {
BASE_URL: 'http://api.openweathermap.org/data/2.5',
DEFAULT_CITY: 'San Francisco',
ENDPOINTS: {
CURRENT: 'weather',
FORECAST: 'forecast'
}
} as const;
class WeatherServer {
private server: Server;
private axiosInstance;
constructor() {
this.server = new Server({
name: "example-weather-server",
version: "0.1.0"
}, {
capabilities: {
resources: {},
tools: {}
}
});
// Configure axios with defaults
this.axiosInstance = axios.create({
baseURL: API_CONFIG.BASE_URL,
params: {
appid: API_KEY,
units: "metric"
}
});
this.setupHandlers();
this.setupErrorHandling();
}
private setupErrorHandling(): void {
this.server.onerror = (error) => {
console.error("[MCP Error]", error);
};
process.on('SIGINT', async () => {
await this.server.close();
process.exit(0);
});
}
private setupHandlers(): void {
this.setupResourceHandlers();
this.setupToolHandlers();
}
private setupResourceHandlers(): void {
// Implementation continues in next section
}
private setupToolHandlers(): void {
// Implementation continues in next section
}
async run(): Promise<void> {
const transport = new StdioServerTransport();
await this.server.connect(transport);
// Although this is just an informative message, we must log to stderr,
// to avoid interfering with MCP communication that happens on stdout
console.error("Weather MCP server running on stdio");
}
}
const server = new WeatherServer();
server.run().catch(console.error);
```
</Step>
<Step title="Add resource handlers">
Add this to the `setupResourceHandlers` method:
```typescript
private setupResourceHandlers(): void {
this.server.setRequestHandler(
ListResourcesRequestSchema,
async () => ({
resources: [{
uri: `weather://${API_CONFIG.DEFAULT_CITY}/current`,
name: `Current weather in ${API_CONFIG.DEFAULT_CITY}`,
mimeType: "application/json",
description: "Real-time weather data including temperature, conditions, humidity, and wind speed"
}]
})
);
this.server.setRequestHandler(
ReadResourceRequestSchema,
async (request) => {
const city = API_CONFIG.DEFAULT_CITY;
if (request.params.uri !== `weather://${city}/current`) {
throw new McpError(
ErrorCode.InvalidRequest,
`Unknown resource: ${request.params.uri}`
);
}
try {
const response = await this.axiosInstance.get<OpenWeatherResponse>(
API_CONFIG.ENDPOINTS.CURRENT,
{
params: { q: city }
}
);
const weatherData: WeatherData = {
temperature: response.data.main.temp,
conditions: response.data.weather[0].description,
humidity: response.data.main.humidity,
wind_speed: response.data.wind.speed,
timestamp: new Date().toISOString()
};
return {
contents: [{
uri: request.params.uri,
mimeType: "application/json",
text: JSON.stringify(weatherData, null, 2)
}]
};
} catch (error) {
if (axios.isAxiosError(error)) {
throw new McpError(
ErrorCode.InternalError,
`Weather API error: ${error.response?.data.message ?? error.message}`
);
}
throw error;
}
}
);
}
```
</Step>
<Step title="Add tool handlers">
Add these handlers to the `setupToolHandlers` method:
```typescript
private setupToolHandlers(): void {
this.server.setRequestHandler(
ListToolsRequestSchema,
async () => ({
tools: [{
name: "get_forecast",
description: "Get weather forecast for a city",
inputSchema: {
type: "object",
properties: {
city: {
type: "string",
description: "City name"
},
days: {
type: "number",
description: "Number of days (1-5)",
minimum: 1,
maximum: 5
}
},
required: ["city"]
}
}]
})
);
this.server.setRequestHandler(
CallToolRequestSchema,
async (request) => {
if (request.params.name !== "get_forecast") {
throw new McpError(
ErrorCode.MethodNotFound,
`Unknown tool: ${request.params.name}`
);
}
if (!isValidForecastArgs(request.params.arguments)) {
throw new McpError(
ErrorCode.InvalidParams,
"Invalid forecast arguments"
);
}
const city = request.params.arguments.city;
const days = Math.min(request.params.arguments.days || 3, 5);
try {
const response = await this.axiosInstance.get<{
list: OpenWeatherResponse[]
}>(API_CONFIG.ENDPOINTS.FORECAST, {
params: {
q: city,
cnt: days * 8 // API returns 3-hour intervals
}
});
const forecasts: ForecastDay[] = [];
for (let i = 0; i < response.data.list.length; i += 8) {
const dayData = response.data.list[i];
forecasts.push({
date: dayData.dt_txt?.split(' ')[0] ?? new Date().toISOString().split('T')[0],
temperature: dayData.main.temp,
conditions: dayData.weather[0].description
});
}
return {
content: [{
type: "text",
text: JSON.stringify(forecasts, null, 2)
}]
};
} catch (error) {
if (axios.isAxiosError(error)) {
return {
content: [{
type: "text",
text: `Weather API error: ${error.response?.data.message ?? error.message}`
}],
isError: true,
}
}
throw error;
}
}
);
}
```
</Step>
<Step title="Build and test">
```bash
npm run build
```
</Step>
</Steps>
## Connect to Claude Desktop
<Steps>
<Step title="Update Claude config">
If you didn't already connect to Claude Desktop during project setup, add to `claude_desktop_config.json`:
```json
{
"mcpServers": {
"weather": {
"command": "node",
"args": ["/path/to/weather-server/build/index.js"],
"env": {
"OPENWEATHER_API_KEY": "your-api-key",
}
}
}
}
```
</Step>
<Step title="Restart Claude">
1. Quit Claude completely
2. Start Claude again
3. Look for your weather server in the 🔌 menu
</Step>
</Steps>
## Try it out!
<AccordionGroup>
<Accordion title="Check Current Weather" active>
Ask Claude:
```
What's the current weather in San Francisco? Can you analyze the conditions?
```
</Accordion>
<Accordion title="Get a Forecast">
Ask Claude:
```
Can you get me a 5-day forecast for Tokyo and tell me if I should pack an umbrella?
```
</Accordion>
<Accordion title="Compare Weather">
Ask Claude:
```
Can you analyze the forecast for both Tokyo and San Francisco and tell me which city will be warmer this week?
```
</Accordion>
</AccordionGroup>
## Understanding the code
<Tabs>
<Tab title="Type Safety">
```typescript
interface WeatherData {
temperature: number;
conditions: string;
humidity: number;
wind_speed: number;
timestamp: string;
}
```
TypeScript adds type safety to our MCP server, making it more reliable and easier to maintain.
</Tab>
<Tab title="Resources">
```typescript
this.server.setRequestHandler(
ListResourcesRequestSchema,
async () => ({
resources: [{
uri: `weather://${DEFAULT_CITY}/current`,
name: `Current weather in ${DEFAULT_CITY}`,
mimeType: "application/json"
}]
})
);
```
Resources provide data that Claude can access as context.
</Tab>
<Tab title="Tools">
```typescript
{
name: "get_forecast",
description: "Get weather forecast for a city",
inputSchema: {
type: "object",
properties: {
city: { type: "string" },
days: { type: "number" }
}
}
}
```
Tools let Claude take actions through your server with type-safe inputs.
</Tab>
</Tabs>
## Best practices
<CardGroup cols={1}>
<Card title="Error Handling" icon="shield">
When a tool encounters an error, return the error message with `isError: true`, so the model can self-correct:
```typescript
try {
const response = await axiosInstance.get(...);
} catch (error) {
if (axios.isAxiosError(error)) {
return {
content: {
mimeType: "text/plain",
text: `Weather API error: ${error.response?.data.message ?? error.message}`
},
isError: true,
}
}
throw error;
}
```
For other handlers, throw an error, so the application can notify the user:
```typescript
try {
const response = await this.axiosInstance.get(...);
} catch (error) {
if (axios.isAxiosError(error)) {
throw new McpError(
ErrorCode.InternalError,
`Weather API error: ${error.response?.data.message}`
);
}
throw error;
}
```
</Card>
<Card title="Type Validation" icon="check">
```typescript
function isValidForecastArgs(args: any): args is GetForecastArgs {
return (
typeof args === "object" &&
args !== null &&
"city" in args &&
typeof args.city === "string"
);
}
```
<Tip>You can also use libraries like [Zod](https://zod.dev/) to perform this validation automatically.</Tip>
</Card>
</CardGroup>
## Available transports
While this guide uses stdio to run the MCP server as a local process, MCP supports other [transports](/docs/concepts/transports) as well.
## Troubleshooting
<Info>
The following troubleshooting tips are for macOS. Guides for other platforms are coming soon.
</Info>
### Build errors
```bash
# Check TypeScript version
npx tsc --version
# Clean and rebuild
rm -rf build/
npm run build
```
### Runtime errors
Look for detailed error messages in the Claude Desktop logs:
```bash
# Monitor logs
tail -n 20 -f ~/Library/Logs/Claude/mcp*.log
```
### Type errors
```bash
# Check types without building
npx tsc --noEmit
```
## Next steps
<CardGroup cols={2}>
<Card title="Architecture overview" icon="sitemap" href="/docs/concepts/architecture">
Learn more about the MCP architecture
</Card>
<Card title="TypeScript SDK" icon="square-js" href="https://github.com/modelcontextprotocol/typescript-sdk">
Check out the TypeScript SDK on GitHub
</Card>
</CardGroup>
<Note>
Need help? Ask Claude! Since it has access to the MCP SDK documentation, it can help you debug issues and suggest improvements to your server.
</Note>
# Debugging
A comprehensive guide to debugging Model Context Protocol (MCP) integrations
Effective debugging is essential when developing MCP servers or integrating them with applications. This guide covers the debugging tools and approaches available in the MCP ecosystem.
<Info>
This guide is for macOS. Guides for other platforms are coming soon.
</Info>
## Debugging tools overview
MCP provides several tools for debugging at different levels:
1. **MCP Inspector**
* Interactive debugging interface
* Direct server testing
* See the [Inspector guide](/docs/tools/inspector) for details
2. **Claude Desktop Developer Tools**
* Integration testing
* Log collection
* Chrome DevTools integration
3. **Server Logging**
* Custom logging implementations
* Error tracking
* Performance monitoring
## Debugging in Claude Desktop
### Checking server status
The Claude.app interface provides basic server status information:
1. Click the 🔌 icon to view:
* Connected servers
* Available prompts and resources
2. Click the 🔨 icon to view:
* Tools made available to the model
### Viewing logs
Review detailed MCP logs from Claude Desktop:
```bash
# Follow logs in real-time
tail -n 20 -f ~/Library/Logs/Claude/mcp*.log
```
The logs capture:
* Server connection events
* Configuration issues
* Runtime errors
* Message exchanges
### Using Chrome DevTools
Access Chrome's developer tools inside Claude Desktop to investigate client-side errors:
1. Enable DevTools:
```bash
jq '.allowDevTools = true' ~/Library/Application\ Support/Claude/developer_settings.json > tmp.json \
&& mv tmp.json ~/Library/Application\ Support/Claude/developer_settings.json
```
2. Open DevTools: `Command-Option-Shift-i`
Note: You'll see two DevTools windows:
* Main content window
* App title bar window
Use the Console panel to inspect client-side errors.
Use the Network panel to inspect:
* Message payloads
* Connection timing
## Common issues
### Environment variables
MCP servers inherit only a subset of environment variables automatically, like `USER`, `HOME`, and `PATH`.
To override the default variables or provide your own, you can specify an `env` key in `claude_desktop_config.json`:
```json
{
"myserver": {
"command": "mcp-server-myapp",
"env": {
"MYAPP_API_KEY": "some_key",
}
}
}
```
### Server initialization
Common initialization problems:
1. **Path Issues**
* Incorrect server executable path
* Missing required files
* Permission problems
2. **Configuration Errors**
* Invalid JSON syntax
* Missing required fields
* Type mismatches
3. **Environment Problems**
* Missing environment variables
* Incorrect variable values
* Permission restrictions
### Connection problems
When servers fail to connect:
1. Check Claude Desktop logs
2. Verify server process is running
3. Test standalone with [Inspector](/docs/tools/inspector)
4. Verify protocol compatibility
## Implementing logging
### Server-side logging
When building a server that uses the local stdio [transport](/docs/concepts/transports), all messages logged to stderr (standard error) will be captured by the host application (e.g., Claude Desktop) automatically.
<Warning>
Local MCP servers should not log messages to stdout (standard out), as this will interfere with protocol operation.
</Warning>
For all [transports](/docs/concepts/transports), you can also provide logging to the client by sending a log message notification:
<Tabs>
<Tab title="Python">
```python
server.request_context.session.send_log_message(
level="info",
data="Server started successfully",
)
```
</Tab>
<Tab title="TypeScript">
```typescript
server.sendLoggingMessage({
level: "info",
data: "Server started successfully",
});
```
</Tab>
</Tabs>
Important events to log:
* Initialization steps
* Resource access
* Tool execution
* Error conditions
* Performance metrics
### Client-side logging
In client applications:
1. Enable debug logging
2. Monitor network traffic
3. Track message exchanges
4. Record error states
## Debugging workflow
### Development cycle
1. Initial Development
* Use [Inspector](/docs/tools/inspector) for basic testing
* Implement core functionality
* Add logging points
2. Integration Testing
* Test in Claude Desktop
* Monitor logs
* Check error handling
### Testing changes
To test changes efficiently:
* **Configuration changes**: Restart Claude Desktop
* **Server code changes**: Use Command-R to reload
* **Quick iteration**: Use [Inspector](/docs/tools/inspector) during development
## Best practices
### Logging strategy
1. **Structured Logging**
* Use consistent formats
* Include context
* Add timestamps
* Track request IDs
2. **Error Handling**
* Log stack traces
* Include error context
* Track error patterns
* Monitor recovery
3. **Performance Tracking**
* Log operation timing
* Monitor resource usage
* Track message sizes
* Measure latency
### Security considerations
When debugging:
1. **Sensitive Data**
* Sanitize logs
* Protect credentials
* Mask personal information
2. **Access Control**
* Verify permissions
* Check authentication
* Monitor access patterns
## Getting help
When encountering issues:
1. **First Steps**
* Check server logs
* Test with [Inspector](/docs/tools/inspector)
* Review configuration
* Verify environment
2. **Support Channels**
* GitHub issues
* GitHub discussions
3. **Providing Information**
* Log excerpts
* Configuration files
* Steps to reproduce
* Environment details
## Next steps
<CardGroup cols={2}>
<Card title="MCP Inspector" icon="magnifying-glass" href="/docs/tools/inspector">
Learn to use the MCP Inspector
</Card>
</CardGroup>
# Inspector
In-depth guide to using the MCP Inspector for testing and debugging Model Context Protocol servers
The [MCP Inspector](https://github.com/modelcontextprotocol/inspector) is an interactive developer tool for testing and debugging MCP servers. While the [Debugging Guide](/docs/tools/debugging) covers the Inspector as part of the overall debugging toolkit, this document provides a detailed exploration of the Inspector's features and capabilities.
## Getting started
### Installation and basic usage
The Inspector runs directly through `npx` without requiring installation:
```bash
npx @modelcontextprotocol/inspector <command>
```
```bash
npx @modelcontextprotocol/inspector <command> <arg1> <arg2>
```
#### Inspecting servers from NPM or PyPi
A common way to start server packages from [NPM](https://npmjs.com) or [PyPi](https://pypi.com).
<Tabs>
<Tab title="NPM package">
```bash
npx -y @modelcontextprotocol/inspector npx <package-name> <args>
# For example
npx -y @modelcontextprotocol/inspector npx server-postgres postgres://127.0.0.1/testdb
```
</Tab>
<Tab title="PyPi package">
```bash
npx @modelcontextprotocol/inspector uvx <package-name> <args>
# For example
npx @modelcontextprotocol/inspector uvx mcp-server-git --repository ~/code/mcp/servers.git
```
</Tab>
</Tabs>
#### Inspecting locally developed servers
To inspect servers locally developed or downloaded as a repository, the most common
way is:
<Tabs>
<Tab title="TypeScript">
```bash
npx @modelcontextprotocol/inspector node path/to/server/index.js args...
```
</Tab>
<Tab title="Python">
```bash
npx @modelcontextprotocol/inspector \
uv \
--directory path/to/server \
run \
package-name \
args...
```
</Tab>
</Tabs>
Please carefully read any attached README for the most accurate instructions.
## Feature overview
<Frame caption="The MCP Inspector interface">
<img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/mcp-inspector.png" />
</Frame>
The Inspector provides several features for interacting with your MCP server:
### Server connection pane
* Allows selecting the [transport](/docs/concepts/transports) for connecting to the server
* For local servers, supports customizing the command-line arguments and environment
### Resources tab
* Lists all available resources
* Shows resource metadata (MIME types, descriptions)
* Allows resource content inspection
* Supports subscription testing
### Prompts tab
* Displays available prompt templates
* Shows prompt arguments and descriptions
* Enables prompt testing with custom arguments
* Previews generated messages
### Tools tab
* Lists available tools
* Shows tool schemas and descriptions
* Enables tool testing with custom inputs
* Displays tool execution results
### Notifications pane
* Presents all logs recorded from the server
* Shows notifications received from the server
## Best practices
### Development workflow
1. Start Development
* Launch Inspector with your server
* Verify basic connectivity
* Check capability negotiation
2. Iterative testing
* Make server changes
* Rebuild the server
* Reconnect the Inspector
* Test affected features
* Monitor messages
3. Test edge cases
* Invalid inputs
* Missing prompt arguments
* Concurrent operations
* Verify error handling and error responses
## Next steps
<CardGroup cols={2}>
<Card title="Inspector Repository" icon="github" href="https://github.com/modelcontextprotocol/inspector">
Check out the MCP Inspector source code
</Card>
<Card title="Debugging Guide" icon="bug" href="/docs/tools/debugging">
Learn about broader debugging strategies
</Card>
</CardGroup>
# Introduction
Get started with the Model Context Protocol (MCP)
The Model Context Protocol (MCP) is an open protocol that enables seamless integration between LLM applications and external data sources and tools. Whether you're building an AI-powered IDE, enhancing a chat interface, or creating custom AI workflows, MCP provides a standardized way to connect LLMs with the context they need.
## Get started with MCP
Choose the path that best fits your needs:
<CardGroup cols={1}>
<Card title="Quickstart" icon="bolt" href="/quickstart">
The fastest way to see MCP in action—connect example servers to Claude Desktop
</Card>
<Card title="Build your first server (Python)" icon="python" href="/docs/first-server/python">
Create a simple MCP server in Python to understand the basics
</Card>
<Card title="Build your first server (TypeScript)" icon="square-js" href="/docs/first-server/typescript">
Create a simple MCP server in TypeScript to understand the basics
</Card>
</CardGroup>
## Development tools
Essential tools for building and debugging MCP servers:
<CardGroup cols={2}>
<Card title="Debugging Guide" icon="bug" href="/docs/tools/debugging">
Learn how to effectively debug MCP servers and integrations
</Card>
<Card title="MCP Inspector" icon="magnifying-glass" href="/docs/tools/inspector">
Test and inspect your MCP servers with our interactive debugging tool
</Card>
</CardGroup>
## Explore MCP
Dive deeper into MCP's core concepts and capabilities:
<CardGroup cols={2}>
<Card title="Core Architecture" icon="sitemap" href="/docs/concepts/architecture">
Understand how MCP connects clients, servers, and LLMs
</Card>
<Card title="Resources" icon="database" href="/docs/concepts/resources">
Expose data and content from your servers to LLMs
</Card>
<Card title="Prompts" icon="message" href="/docs/concepts/prompts">
Create reusable prompt templates and workflows
</Card>
<Card title="Tools" icon="wrench" href="/docs/concepts/tools">
Enable LLMs to perform actions through your server
</Card>
<Card title="Sampling" icon="robot" href="/docs/concepts/sampling">
Let your servers request completions from LLMs
</Card>
<Card title="Transports" icon="network-wired" href="/docs/concepts/transports">
Learn about MCP's communication mechanism
</Card>
</CardGroup>
## Contributing
Want to contribute? Check out [@modelcontextprotocol](https://github.com/modelcontextprotocol) on GitHub to join our growing community of developers building with MCP.
# Quickstart
Get started with MCP in less than 5 minutes
MCP is a protocol that enables secure connections between host applications, such as [Claude Desktop](https://claude.ai/download), and local services. In this quickstart guide, you'll learn how to:
* Set up a local SQLite database
* Connect Claude Desktop to it through MCP
* Query and analyze your data securely
<Note>
While this guide focuses on using Claude Desktop as an example MCP host, the protocol is open and can be integrated by any application. IDEs, AI tools, and other software can all use MCP to connect to local integrations in a standardized way.
</Note>
<Warning>
Claude Desktop's MCP support is currently in developer preview and only supports connecting to local MCP servers running on your machine. Remote MCP connections are not yet supported. This integration is only available in the Claude Desktop app, not the Claude web interface (claude.ai).
</Warning>
## How MCP works
MCP (Model Context Protocol) is an open protocol that enables secure, controlled interactions between AI applications and local or remote resources. Let's break down how it works, then look at how we'll use it in this guide.
### General Architecture
At its core, MCP follows a client-server architecture where a host application can connect to multiple servers:
```mermaid
flowchart LR
subgraph "Your Computer"
Host["MCP Host\n(Claude, IDEs, Tools)"]
S1["MCP Server A"]
S2["MCP Server B"]
S3["MCP Server C"]
Host <-->|"MCP Protocol"| S1
Host <-->|"MCP Protocol"| S2
Host <-->|"MCP Protocol"| S3
S1 <--> R1[("Local\nResource A")]
S2 <--> R2[("Local\nResource B")]
end
subgraph "Internet"
S3 <-->|"Web APIs"| R3[("Remote\nResource C")]
end
```
* **MCP Hosts**: Programs like Claude Desktop, IDEs, or AI tools that want to access resources through MCP
* **MCP Clients**: Protocol clients that maintain 1:1 connections with servers
* **MCP Servers**: Lightweight programs that each expose specific capabilities through the standardized Model Context Protocol
* **Local Resources**: Your computer's resources (databases, files, services) that MCP servers can securely access
* **Remote Resources**: Resources available over the internet (e.g., through APIs) that MCP servers can connect to
### In This Guide
For this quickstart, we'll implement a focused example using SQLite:
```mermaid
flowchart LR
subgraph "Your Computer"
direction LR
Claude["Claude Desktop"]
MCP["SQLite MCP Server"]
DB[(SQLite Database\n~/test.db)]
Claude <-->|"MCP Protocol\n(Queries & Results)"| MCP
MCP <-->|"Local Access\n(SQL Operations)"| DB
end
```
1. Claude Desktop acts as our MCP client
2. A SQLite MCP Server provides secure database access
3. Your local SQLite database stores the actual data
The communication between the SQLite MCP server and your local SQLite database happens entirely on your machine—your SQLite database is not exposed to the internet. The Model Context Protocol ensures that Claude Desktop can only perform approved database operations through well-defined interfaces. This gives you a secure way to let Claude analyze and interact with your local data while maintaining complete control over what it can access.
## Prerequisites
* macOS or Windows
* The latest version of [Claude Desktop](https://claude.ai/download) installed
* [uv](https://docs.astral.sh/uv/) 0.4.18 or higher (`uv --version` to check)
* Git (`git --version` to check)
* SQLite (`sqlite3 --version` to check)
<AccordionGroup>
<Accordion title="Installing prerequisites (macOS)">
```bash
# Using Homebrew
brew install uv git sqlite3
# Or download directly:
# uv: https://docs.astral.sh/uv/
# Git: https://git-scm.com
# SQLite: https://www.sqlite.org/download.html
```
</Accordion>
<Accordion title="Installing prerequisites (Windows)">
```powershell
# Using winget
winget install --id=astral-sh.uv -e
winget install git.git sqlite.sqlite
# Or download directly:
# uv: https://docs.astral.sh/uv/
# Git: https://git-scm.com
# SQLite: https://www.sqlite.org/download.html
```
</Accordion>
</AccordionGroup>
## Installation
<Tabs>
<Tab title="macOS">
<Steps>
<Step title="Create a sample database">
Let's create a simple SQLite database for testing:
```bash
# Create a new SQLite database
sqlite3 ~/test.db <<EOF
CREATE TABLE products (
id INTEGER PRIMARY KEY,
name TEXT,
price REAL
);
INSERT INTO products (name, price) VALUES
('Widget', 19.99),
('Gadget', 29.99),
('Gizmo', 39.99),
('Smart Watch', 199.99),
('Wireless Earbuds', 89.99),
('Portable Charger', 24.99),
('Bluetooth Speaker', 79.99),
('Phone Stand', 15.99),
('Laptop Sleeve', 34.99),
('Mini Drone', 299.99),
('LED Desk Lamp', 45.99),
('Keyboard', 129.99),
('Mouse Pad', 12.99),
('USB Hub', 49.99),
('Webcam', 69.99),
('Screen Protector', 9.99),
('Travel Adapter', 27.99),
('Gaming Headset', 159.99),
('Fitness Tracker', 119.99),
('Portable SSD', 179.99);
EOF
```
</Step>
<Step title="Configure Claude Desktop">
Open your Claude Desktop App configuration at `~/Library/Application Support/Claude/claude_desktop_config.json` in a text editor.
For example, if you have [VS Code](https://code.visualstudio.com/) installed:
```bash
code ~/Library/Application\ Support/Claude/claude_desktop_config.json
```
Add this configuration (replace YOUR\_USERNAME with your actual username):
```json
{
"mcpServers": {
"sqlite": {
"command": "uvx",
"args": ["mcp-server-sqlite", "--db-path", "/Users/YOUR_USERNAME/test.db"]
}
}
}
```
This tells Claude Desktop:
1. There's an MCP server named "sqlite"
2. Launch it by running `uvx mcp-server-sqlite`
3. Connect it to your test database
Save the file, and restart **Claude Desktop**.
</Step>
</Steps>
</Tab>
<Tab title="Windows">
<Steps>
<Step title="Create a sample database">
Let's create a simple SQLite database for testing:
```powershell
# Create a new SQLite database
$sql = @'
CREATE TABLE products (
id INTEGER PRIMARY KEY,
name TEXT,
price REAL
);
INSERT INTO products (name, price) VALUES
('Widget', 19.99),
('Gadget', 29.99),
('Gizmo', 39.99),
('Smart Watch', 199.99),
('Wireless Earbuds', 89.99),
('Portable Charger', 24.99),
('Bluetooth Speaker', 79.99),
('Phone Stand', 15.99),
('Laptop Sleeve', 34.99),
('Mini Drone', 299.99),
('LED Desk Lamp', 45.99),
('Keyboard', 129.99),
('Mouse Pad', 12.99),
('USB Hub', 49.99),
('Webcam', 69.99),
('Screen Protector', 9.99),
('Travel Adapter', 27.99),
('Gaming Headset', 159.99),
('Fitness Tracker', 119.99),
('Portable SSD', 179.99);
'@
cd ~
& sqlite3 test.db $sql
```
</Step>
<Step title="Configure Claude Desktop">
Open your Claude Desktop App configuration at `%APPDATA%\Claude\claude_desktop_config.json` in a text editor.
For example, if you have [VS Code](https://code.visualstudio.com/) installed:
```powershell
code $env:AppData\Claude\claude_desktop_config.json
```
Add this configuration (replace YOUR\_USERNAME with your actual username):
```json
{
"mcpServers": {
"sqlite": {
"command": "uvx",
"args": [
"mcp-server-sqlite",
"--db-path",
"C:\\Users\\YOUR_USERNAME\\test.db"
]
}
}
}
```
This tells Claude Desktop:
1. There's an MCP server named "sqlite"
2. Launch it by running `uvx mcp-server-sqlite`
3. Connect it to your test database
Save the file, and restart **Claude Desktop**.
</Step>
</Steps>
</Tab>
</Tabs>
## Test it out
Let's verify everything is working. Try sending this prompt to Claude Desktop:
```
Can you connect to my SQLite database and tell me what products are available, and their prices?
```
Claude Desktop will:
1. Connect to the SQLite MCP server
2. Query your local database
3. Format and present the results
<Frame caption="Claude Desktop successfully queries our SQLite database 🎉">
<img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-screenshot.png" alt="Example Claude Desktop conversation showing database query results" />
</Frame>
## What's happening under the hood?
When you interact with Claude Desktop using MCP:
1. **Server Discovery**: Claude Desktop connects to your configured MCP servers on startup
2. **Protocol Handshake**: When you ask about data, Claude Desktop:
* Identifies which MCP server can help (sqlite in this case)
* Negotiates capabilities through the protocol
* Requests data or actions from the MCP server
3. **Interaction Flow**:
```mermaid
sequenceDiagram
participant C as Claude Desktop
participant M as MCP Server
participant D as SQLite DB
C->>M: Initialize connection
M-->>C: Available capabilities
C->>M: Query request
M->>D: SQL query
D-->>M: Results
M-->>C: Formatted results
```
4. **Security**:
* MCP servers only expose specific, controlled capabilities
* MCP servers run locally on your machine, and the resources they access are not exposed to the internet
* Claude Desktop requires user confirmation for sensitive operations
## Try these examples
Now that MCP is working, try these increasingly powerful examples:
<AccordionGroup>
<Accordion title="Basic Queries" active>
```
What's the average price of all products in the database?
```
</Accordion>
<Accordion title="Data Analysis">
```
Can you analyze the price distribution and suggest any pricing optimizations?
```
</Accordion>
<Accordion title="Complex Operations">
```
Could you help me design and create a new table for storing customer orders?
```
</Accordion>
</AccordionGroup>
## Add more capabilities
Want to give Claude Desktop more local integration capabilities? Add these servers to your configuration:
<Note>
Note that these MCP servers will require [Node.js](https://nodejs.org/en) to be installed on your machine.
</Note>
<AccordionGroup>
<Accordion title="File System Access" icon="folder-open">
Add this to your config to let Claude Desktop read and analyze files:
```json
"filesystem": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/YOUR_USERNAME/Desktop"]
}
```
</Accordion>
<Accordion title="PostgreSQL Connection" icon="database">
Connect Claude Desktop to your PostgreSQL database:
```json
"postgres": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://localhost/mydb"]
}
```
</Accordion>
</AccordionGroup>
## More MCP Clients
While this guide demonstrates MCP using Claude Desktop as a client, several other applications support MCP integration:
<CardGroup cols={2}>
<Card title="Zed Editor" icon="pen-to-square" href="https://zed.dev">
A high-performance, multiplayer code editor with built-in MCP support for AI-powered coding assistance
</Card>
<Card title="Cody" icon="magnifying-glass" href="https://sourcegraph.com/cody">
Code intelligence platform featuring MCP integration for enhanced code search and analysis capabilities
</Card>
</CardGroup>
Each host application may implement MCP features differently or support different capabilities. Check their respective documentation for specific setup instructions and supported features.
## Troubleshooting
<AccordionGroup>
<Accordion title="Nothing showing up in Claude Desktop?">
1. Check if MCP is enabled:
* Click the 🔌 icon in Claude Desktop, next to the chat box
* Expand "Installed MCP Servers"
* You should see your configured servers
2. Verify your config:
* From Claude Desktop, go to Claude > Settings…
* Open the "Developer" tab to see your configuration
3. Restart Claude Desktop completely:
* Quit the app (not just close the window)
* Start it again
</Accordion>
<Accordion title="MCP or database errors?">
1. Check Claude Desktop's logs:
```bash
tail -n 20 -f ~/Library/Logs/Claude/mcp*.log
```
2. Verify database access:
```bash
# Test database connection
sqlite3 ~/test.db ".tables"
```
3. Common fixes:
* Check file paths in your config
* Verify database file permissions
* Ensure SQLite is installed properly
</Accordion>
</AccordionGroup>
## Next steps
<CardGroup cols={2}>
<Card title="Build your first MCP server" icon="code" href="/docs/first-server/python">
Create your own MCP servers to give your LLM clients new capabilities.
</Card>
<Card title="Explore examples" icon="github" href="https://github.com/modelcontextprotocol/servers">
Browse our collection of example servers to see what's possible.
</Card>
</CardGroup>
```