mcp-integration-expert（MCP 集成 expert）

MCP Integration Expert

Comprehensive skill for researching, documenting, and integrating Model Context Protocol (MCP) servers and tools into Claude Code and AI applications.

When to Use This Skill

Use this skill when you need to:

• Research and evaluate MCP servers for integration
• Build custom MCP servers or clients
• Integrate existing MCP tools into Claude Code
• Document MCP server capabilities and usage patterns
• Troubleshoot MCP integration issues
• Implement MCP security and authentication patterns
• Create multi-language MCP implementations (Python, TypeScript, C#, Java, Rust)
• Design MCP-based AI agent workflows
• Evaluate MCP server trust scores and documentation quality

Model Context Protocol (MCP) Overview

What is MCP?

The Model Context Protocol (MCP) is an open standard introduced by Anthropic in November 2024 that standardizes how AI applications and Large Language Models (LLMs) integrate with external data sources, tools, and systems.

Key Analogy: MCP is like a "USB-C port for AI" - providing a universal, standardized interface for connecting AI models to diverse data sources and tools.

Core Architecture

┌─────────────────────────────────────────────────────────────┐
│                        MCP Client                           │
│                     (AI Application)                        │
│                                                             │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐    │
│  │   Claude     │  │   ChatGPT    │  │    Custom    │    │
│  │     Code     │  │   Desktop    │  │      App     │    │
│  └──────────────┘  └──────────────┘  └──────────────┘    │
└─────────────────────────────────────────────────────────────┘
                          │
                          │ MCP Protocol
                          │
┌─────────────────────────────────────────────────────────────┐
│                      MCP Servers                            │
│                                                             │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐  ┌──────────┐  │
│  │  Linear  │  │ Postgres │  │  GitHub  │  │  Custom  │  │
│  │  Server  │  │  Server  │  │  Server  │  │  Server  │  │
│  └──────────┘  └──────────┘  └──────────┘  └──────────┘  │
└─────────────────────────────────────────────────────────────┘

MCP Protocol Communication Flow

sequenceDiagram
    autonumber
    actor User as 👤 User
    participant ClientApp as 🖥️ Client App
    participant ClientLLM as 🧠 Client LLM
    participant Server1 as 🔧 MCP Server 1
    participant Server2 as 📚 MCP Server 2

    %% Discovery Phase
    rect rgb(220, 240, 255)
        Note over ClientApp, Server2: TOOL DISCOVERY PHASE
        ClientApp->>+Server1: Request available tools/resources
        Server1-->>-ClientApp: Return tool list (JSON)
        ClientApp->>+Server2: Request available tools/resources
        Server2-->>-ClientApp: Return tool list (JSON)
        Note right of ClientApp: Store combined tool<br/>catalog locally
    end

    %% User Interaction
    rect rgb(255, 240, 220)
        Note over User, ClientLLM: USER INTERACTION PHASE
        User->>+ClientApp: Enter natural language prompt
        ClientApp->>+ClientLLM: Forward prompt + tool catalog
        ClientLLM->>-ClientLLM: Analyze prompt & select tools
    end

    %% Tool Execution
    rect rgb(220, 255, 220)
        Note over ClientApp, Server1: TOOL EXECUTION PHASE
        ClientLLM->>+ClientApp: Request tool execution
        ClientApp->>+Server1: Execute specific tool
        Server1-->>-ClientApp: Return results
        ClientApp->>+ClientLLM: Process results
        ClientLLM-->>-ClientApp: Generate response
        ClientApp-->>-User: Display final answer
    end

MCP Core Primitives

MCP provides three core primitives for context exchange:

1. Resources: Expose data sources (files, databases, APIs)
2. Tools: Enable actions and operations
3. Prompts: Provide reusable prompt templates

MCP Integration Workflow

Phase 1: Research & Discovery

1.1 Identify MCP Server Needs

Questions to ask:

• What data sources do you need to access?
• What actions/tools do you need to perform?
• Are there existing MCP servers for your use case?
• Do you need to build a custom MCP server?

1.2 Research Available MCP Servers

Official MCP Server Repository: https://github.com/modelcontextprotocol

Popular MCP Servers (2025):

• Linear MCP: Project management and issue tracking
• Playwright MCP: Browser automation and testing
• Context7 MCP: Library documentation retrieval
• GitHub MCP: Repository management and automation
• Postgres MCP: Database queries and operations
• Google Drive MCP: File storage and retrieval
• Slack MCP: Team communication
• Stripe MCP: Payment processing
• Puppeteer MCP: Web scraping and automation

1.3 Evaluate MCP Server Quality

Trust Score Criteria (1-10 scale):

• 9-10: Highly authoritative (official SDKs, major platforms)
• 7-9: Well-maintained community projects
• 5-7: Experimental or niche implementations
• <5: Proof-of-concept or unmaintained

Documentation Coverage:

• High: 500+ code snippets
• Medium: 100-500 code snippets
• Low: <100 code snippets

Phase 2: MCP Server Architecture

2.1 MCP Server Components

// TypeScript MCP Server Structure
import { FastMCP } from '@modelcontextprotocol/typescript-sdk';

const mcp = new FastMCP({
    name: "My MCP Server",
    version: "1.0.0"
});

// 1. Resources: Expose data
mcp.resource("greeting://{name}", (name: string) => {
    return `Hello, ${name}!`;
});

// 2. Tools: Enable actions
mcp.tool("calculate", {
    description: "Perform calculations",
    parameters: {
        operation: { type: "string" },
        a: { type: "number" },
        b: { type: "number" }
    }
}, async (params) => {
    // Tool implementation
});

// 3. Prompts: Reusable templates
mcp.prompt("code-review", {
    description: "Code review template",
    arguments: ["language", "code"]
});

2.2 Transport Mechanisms

MCP supports multiple transport mechanisms:

1. Stdio Transport (Default for local servers)

from mcp.server.fastmcp import FastMCP
from mcp.transports.stdio import serve_stdio

mcp = FastMCP("Demo")

# Run with stdio
asyncio.run(serve_stdio(mcp))

2. HTTP/SSE Transport (For remote servers)

import { StreamableHTTPServerTransport } from '@modelcontextprotocol/typescript-sdk';

const transport = new StreamableHTTPServerTransport(8000);
await server.connect(transport);

Phase 3: MCP Client Integration

3.1 Client Setup (Python)

from mcp.client import stdio_client, ClientSession

async def run_client():
    server_params = {
        "command": "python",
        "args": ["server.py"]
    }

    async with stdio_client(server_params) as (reader, writer):
        async with ClientSession(reader, writer) as session:
            # Initialize connection
            await session.initialize()

            # Discover tools
            tools = await session.list_tools()

            # Call a tool
            result = await session.call_tool("add", arguments={"a": 5, "b": 7})
            print(f"Result: {result}")

3.2 Client Setup (TypeScript)

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StdioClientTransport } from "@modelcontextprotocol/sdk/client/stdio.js";

class MCPClient {
    private client: Client;

    constructor() {
        this.client = new Client({
            name: "example-client",
            version: "1.0.0"
        }, {
            capabilities: {
                prompts: {},
                resources: {},
                tools: {}
            }
        });
    }

    async connectToServer(transport: Transport) {
        await this.client.connect(transport);
    }

    async listTools() {
        return await this.client.listTools();
    }

    async callTool(name: string, args: any) {
        return await this.client.callTool({
            name: name,
            arguments: args
        });
    }
}

3.3 Client Setup (C#)

using ModelContextProtocol.Client;
using ModelContextProtocol.Protocol.Transport;

var clientTransport = new StdioClientTransport(new()
{
    Name = "Demo Server",
    Command = "/path/to/server/executable",
    Arguments = [],
});

await using var mcpClient = await McpClientFactory.CreateAsync(clientTransport);

// List tools
var tools = await mcpClient.ListToolsAsync();

foreach (var tool in tools)
{
    Console.WriteLine($"Tool: {tool.Name}");
    Console.WriteLine($"Description: {tool.Description}");
}

// Call a tool
var result = await mcpClient.CallToolAsync("add", new { a = 5, b = 7 });

Phase 4: Building Custom MCP Servers

4.1 Python MCP Server (FastMCP)

from fastmcp import FastMCP
from fastmcp.transports.stdio import serve_stdio
import asyncio

mcp = FastMCP(
    name="Weather MCP Server",
    version="1.0.0"
)

@mcp.tool()
def get_weather(location: str) -> dict:
    """Gets current weather for a location."""
    # Implementation would call weather API
    return {
        "temperature": 72.5,
        "conditions": "Sunny",
        "location": location
    }

@mcp.resource("weather://{location}")
def weather_resource(location: str) -> str:
    """Get weather data as a resource"""
    return f"Current weather in {location}"

class WeatherTools:
    @mcp.tool()
    def forecast(self, location: str, days: int = 1) -> dict:
        """Gets weather forecast"""
        return {
            "location": location,
            "forecast": [
                {"day": i+1, "temperature": 70 + i, "conditions": "Partly Cloudy"}
                for i in range(days)
            ]
        }

weather_tools = WeatherTools()

if __name__ == "__main__":
    asyncio.run(serve_stdio(mcp))

4.2 TypeScript MCP Server

import { FastMCP } from '@modelcontextprotocol/typescript-sdk';
import { serve_stdio } from '@modelcontextprotocol/typescript-sdk/transports/stdio';

const mcp = new FastMCP({
    name: "Weather MCP Server",
    version: "1.0.0"
});

mcp.tool("get_weather", {
    description: "Gets current weather for a location",
    parameters: {
        location: { type: "string", required: true }
    }
}, async (params) => {
    return {
        temperature: 72.5,
        conditions: "Sunny",
        location: params.location
    };
});

mcp.resource("weather://{location}", async (location: string) => {
    return `Current weather in ${location}`;
});

// Start server
serve_stdio(mcp);

4.3 Java MCP Server

import io.modelcontextprotocol.server.McpServer;
import io.modelcontextprotocol.server.McpToolDefinition;
import io.modelcontextprotocol.server.transport.StdioServerTransport;

public class WeatherMcpServer {
    public static void main(String[] args) throws Exception {
        McpServer server = McpServer.builder()
            .name("Weather MCP Server")
            .version("1.0.0")
            .build();

        server.registerTool(McpToolDefinition.builder("weatherTool")
            .description("Gets current weather for a location")
            .parameter("location", String.class)
            .execute((ctx) -> {
                String location = ctx.getParameter("location", String.class);
                WeatherData data = getWeatherData(location);

                return ToolResponse.content(
                    String.format("Temperature: %.1f°F, Conditions: %s",
                        data.getTemperature(),
                        data.getConditions())
                );
            })
            .build());

        try (StdioServerTransport transport = new StdioServerTransport()) {
            server.connect(transport);
            Thread.currentThread().join();
        }
    }
}

Phase 5: LLM Integration Patterns

5.1 Integrating MCP with OpenAI

import OpenAI from "openai";
import { Client } from "@modelcontextprotocol/sdk/client/index.js";

class MCPOpenAIClient {
    private openai: OpenAI;
    private mcpClient: Client;

    constructor() {
        this.openai = new OpenAI({
            apiKey: process.env.OPENAI_API_KEY,
        });

        this.mcpClient = new Client({
            name: "openai-mcp-client",
            version: "1.0.0"
        }, {
            capabilities: { tools: {} }
        });
    }

    // Adapt MCP tools to OpenAI format
    mcpToolToOpenAITool(tool: any) {
        return {
            type: "function" as const,
            function: {
                name: tool.name,
                description: tool.description,
                parameters: {
                    type: "object",
                    properties: tool.input_schema.properties,
                    required: tool.input_schema.required,
                },
            },
        };
    }

    async processRequest(userMessage: string) {
        // Get MCP tools
        const mcpTools = await this.mcpClient.listTools();
        const openaiTools = mcpTools.map(t => this.mcpToolToOpenAITool(t));

        // Call OpenAI with tools
        const response = await this.openai.chat.completions.create({
            model: "gpt-4",
            messages: [{ role: "user", content: userMessage }],
            tools: openaiTools,
        });

        // Handle tool calls
        const toolCalls = response.choices[0].message.tool_calls;
        if (toolCalls) {
            for (const toolCall of toolCalls) {
                const result = await this.mcpClient.callTool({
                    name: toolCall.function.name,
                    arguments: JSON.parse(toolCall.function.arguments),
                });
                console.log("Tool result:", result);
            }
        }
    }
}

5.2 Integrating MCP with Azure OpenAI (C#)

using Azure.AI.Inference;
using Azure;
using ModelContextProtocol.Client;
using System.Text.Json;

var endpoint = "https://models.inference.ai.azure.com";
var token = Environment.GetEnvironmentVariable("GITHUB_TOKEN");
var client = new ChatCompletionsClient(new Uri(endpoint), new AzureKeyCredential(token));

var mcpClient = await McpClientFactory.CreateAsync(clientTransport);

// Convert MCP tools to Azure format
ChatCompletionsToolDefinition ConvertFrom(string name, string description, JsonElement schema)
{
    FunctionDefinition functionDefinition = new FunctionDefinition(name)
    {
        Description = description,
        Parameters = BinaryData.FromObjectAsJson(new
        {
            Type = "object",
            Properties = schema
        })
    };

    return new ChatCompletionsToolDefinition(functionDefinition);
}

// Get tools from MCP server
var mcpTools = await mcpClient.ListToolsAsync();
var toolDefinitions = new List<ChatCompletionsToolDefinition>();

foreach (var tool in mcpTools)
{
    JsonElement propertiesElement;
    tool.JsonSchema.TryGetProperty("properties", out propertiesElement);

    var def = ConvertFrom(tool.Name, tool.Description, propertiesElement);
    toolDefinitions.Add(def);
}

// Use tools in chat completion
var chatHistory = new List<ChatRequestMessage>
{
    new ChatRequestSystemMessage("You are a helpful assistant"),
    new ChatRequestUserMessage("What's the weather in Seattle?")
};

var response = await client.CompleteAsync(chatHistory,
    new ChatCompletionsOptions { Tools = toolDefinitions });

Phase 6: MCP Configuration in Claude Code

6.1 Claude Code MCP Configuration

Claude Code automatically discovers and loads MCP servers configured in claude_desktop_config.json:

Location:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

Configuration Format:

{
  "mcpServers": {
    "linear": {
      "command": "npx",
      "args": ["-y", "@linear/mcp-server"],
      "env": {
        "LINEAR_API_KEY": "your-api-key"
      }
    },
    "playwright": {
      "command": "npx",
      "args": ["-y", "@playwright/mcp-server"]
    },
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres"],
      "env": {
        "POSTGRES_CONNECTION_STRING": "postgresql://user:pass@localhost/db"
      }
    },
    "custom-server": {
      "command": "python",
      "args": ["/path/to/custom/server.py"]
    }
  }
}

6.2 Verify MCP Server Connection

Once configured, MCP servers automatically connect when Claude Code starts.

Available MCP tools in Claude Code:

// All MCP tools are prefixed with "mcp__<server-name>__<tool-name>"
mcp__linear__create_issue
mcp__linear__list_issues
mcp__linear__update_issue
mcp__playwright__browser_navigate
mcp__playwright__browser_screenshot
mcp__context7__resolve_library_id
mcp__context7__get_library_docs

Phase 7: Security & Best Practices

7.1 Security Considerations (2025 Update)

Known Security Issues (April 2025):

• Prompt Injection: MCP tools can be exploited via prompt injection attacks
• Tool Permissions: Combining tools can enable unintended file exfiltration
• Lookalike Tools: Malicious tools can silently replace trusted ones
• Credential Exposure: API keys and credentials must be protected

Security Best Practices:

1. Validate Tool Inputs

@mcp.tool()
def read_file(filepath: str) -> str:
    """Read a file - with security validation"""
    # Validate filepath to prevent path traversal
    import os
    filepath = os.path.normpath(filepath)

    # Ensure within allowed directory
    allowed_dir = "/home/user/documents"
    if not filepath.startswith(allowed_dir):
        raise ValueError("Access denied: Path outside allowed directory")

    with open(filepath, 'r') as f:
        return f.read()

1. Implement Authentication

from fastmcp import FastMCP
import os

mcp = FastMCP("Secure Server", version="1.0.0")

@mcp.middleware
async def authenticate(request, call_next):
    api_key = request.headers.get("X-API-Key")
    expected_key = os.getenv("MCP_API_KEY")

    if api_key != expected_key:
        raise PermissionError("Invalid API key")

    return await call_next(request)

1. Rate Limiting

from collections import defaultdict
import time

call_counts = defaultdict(list)

@mcp.middleware
async def rate_limit(request, call_next):
    client_id = request.client_id
    now = time.time()

    # Clean old entries
    call_counts[client_id] = [t for t in call_counts[client_id] if now - t < 60]

    # Check rate limit (10 calls per minute)
    if len(call_counts[client_id]) >= 10:
        raise Exception("Rate limit exceeded")

    call_counts[client_id].append(now)
    return await call_next(request)

7.2 Error Handling Best Practices

from fastmcp import FastMCP
from typing import Optional
import logging

mcp = FastMCP("Robust Server", version="1.0.0")
logger = logging.getLogger(__name__)

@mcp.tool()
def safe_api_call(endpoint: str, params: Optional[dict] = None) -> dict:
    """Make an API call with comprehensive error handling"""
    try:
        # Validate inputs
        if not endpoint.startswith("https://"):
            raise ValueError("Only HTTPS endpoints allowed")

        # Make API call
        response = requests.get(endpoint, params=params, timeout=5)
        response.raise_for_status()

        return {
            "success": True,
            "data": response.json()
        }

    except requests.Timeout:
        logger.error(f"Timeout calling {endpoint}")
        return {
            "success": False,
            "error": "Request timeout",
            "error_type": "timeout"
        }

    except requests.HTTPError as e:
        logger.error(f"HTTP error calling {endpoint}: {e}")
        return {
            "success": False,
            "error": str(e),
            "error_type": "http_error",
            "status_code": e.response.status_code
        }

    except Exception as e:
        logger.exception(f"Unexpected error calling {endpoint}")
        return {
            "success": False,
            "error": "Internal server error",
            "error_type": "internal_error"
        }

7.3 Testing MCP Servers

Integration Testing (Python):

import pytest
from mcp.server import McpServer
from mcp.client import McpClient

@pytest.mark.asyncio
async def test_mcp_server_integration():
    # Start test server
    server = McpServer()
    server.register_tool(WeatherForecastTool(MockWeatherService()))
    await server.start(port=5000)

    try:
        # Create client
        client = McpClient("http://localhost:5000")

        # Test tool discovery
        tools = await client.discover_tools()
        assert "weatherForecast" in [t.name for t in tools]

        # Test tool execution
        response = await client.execute_tool("weatherForecast", {
            "location": "Seattle",
            "days": 3
        })

        # Verify response
        assert response.status_code == 200
        assert "Seattle" in response.content[0].text

    finally:
        await server.stop()

Phase 8: Advanced MCP Patterns

8.1 Chain of Tools Workflow

class ChainWorkflow:
    def __init__(self, tools_chain):
        self.tools_chain = tools_chain

    async def execute(self, mcp_client, initial_input):
        current_result = initial_input
        all_results = {"input": initial_input}

        for tool_name in self.tools_chain:
            response = await mcp_client.execute_tool(tool_name, current_result)
            all_results[tool_name] = response.result
            current_result = response.result

        return {
            "final_result": current_result,
            "all_results": all_results
        }

# Usage
data_pipeline = ChainWorkflow([
    "dataFetch",
    "dataCleaner",
    "dataAnalyzer",
    "dataVisualizer"
])

result = await data_pipeline.execute(
    mcp_client,
    {"source": "sales_database", "table": "transactions"}
)

8.2 Parallel Tool Execution

async function executeParallelTools(mcpClient: Client, tools: ToolCall[]) {
    const promises = tools.map(tool =>
        mcpClient.callTool({
            name: tool.name,
            arguments: tool.arguments
        })
    );

    const results = await Promise.all(promises);
    return results;
}

// Usage
const toolCalls = [
    { name: "get_weather", arguments: { location: "Seattle" } },
    { name: "get_weather", arguments: { location: "Portland" } },
    { name: "get_weather", arguments: { location: "Vancouver" } }
];

const weatherData = await executeParallelTools(mcpClient, toolCalls);

8.3 Context-Aware Tool Selection

from typing import List, Dict
import openai

class ContextAwareToolSelector:
    def __init__(self, mcp_client, llm_client):
        self.mcp_client = mcp_client
        self.llm_client = llm_client

    async def select_tools(self, user_query: str, available_tools: List[Dict]) -> List[str]:
        """Use LLM to intelligently select which tools to use"""

        tool_descriptions = "\n".join([
            f"- {tool['name']}: {tool['description']}"
            for tool in available_tools
        ])

        prompt = f"""
        User query: {user_query}

        Available tools:
        {tool_descriptions}

        Select the most relevant tools to answer this query.
        Return a JSON array of tool names.
        """

        response = await self.llm_client.chat.completions.create(
            model="gpt-4",
            messages=[{"role": "user", "content": prompt}],
            response_format={"type": "json_object"}
        )

        selected_tools = json.loads(response.choices[0].message.content)
        return selected_tools["tools"]

MCP Ecosystem (2025)

Major Platform Adoptions

OpenAI (March 2025):

• MCP integrated into ChatGPT Desktop
• MCP support in Agents SDK
• MCP compatibility in Responses API

Google (April 2025):

• MCP support in Gemini models
• Data Commons MCP Server (public datasets)
• MCP integration in Google DeepMind infrastructure

Microsoft (2025):

• MCP in Copilot Studio (GA)
• Semantic Kernel integration
• Azure OpenAI compatibility

Anthropic:

• Native MCP support in Claude Code
• Reference MCP server implementations
• MCP specification maintenance

Official MCP Servers

Repository: https://github.com/modelcontextprotocol

Enterprise Integrations:

• Google Drive, Slack, GitHub, GitLab
• Postgres, MySQL, MongoDB
• AWS, Azure, GCP
• Stripe, Salesforce

Development Tools:

• Puppeteer, Playwright
• Docker, Kubernetes
• Git, GitHub Actions

MCP SDK Support (2025)

Official SDKs:

• Python: pip install modelcontextprotocol
• TypeScript: npm install @modelcontextprotocol/sdk
• C#: NuGet package
• Java: Maven/Gradle
• Rust: Cargo package
• Swift: Swift Package Manager

Documentation Research Workflow

Using Context7 for MCP Research

When researching MCP servers and integration patterns:

# 1. Resolve library ID
/ctx7 model context protocol

# 2. Get comprehensive documentation
# Context7 returns documentation with:
# - Trust Score (7-10 for quality sources)
# - Code Snippets (100-5000+)
# - Implementation examples
# - Best practices

# 3. Evaluate quality
# - Trust Score 9-10: Official documentation
# - Trust Score 7-9: Well-maintained projects
# - Code Snippets 500+: Comprehensive coverage

Best Context7 Sources for MCP (by Trust Score):

1. /microsoft/mcp-for-beginners (Trust: 9.9, Snippets: 30,945)
2. /modelcontextprotocol/python-sdk (Trust: 7.8, Snippets: 119)
3. /modelcontextprotocol/typescript-sdk (Trust: 7.8, Snippets: 55)
4. /modelcontextprotocol/csharp-sdk (Trust: 7.8, Snippets: 59)

Common MCP Integration Patterns

Pattern 1: Simple Tool Integration

Use Case: Add a single MCP tool to Claude Code

// claude_desktop_config.json
{
  "mcpServers": {
    "weather": {
      "command": "python",
      "args": ["weather_server.py"]
    }
  }
}

# weather_server.py
from fastmcp import FastMCP
from fastmcp.transports.stdio import serve_stdio
import asyncio

mcp = FastMCP("Weather", version="1.0.0")

@mcp.tool()
def get_weather(location: str) -> str:
    return f"Weather in {location}: Sunny, 72°F"

asyncio.run(serve_stdio(mcp))

Pattern 2: Multi-Server Integration

Use Case: Combine multiple MCP servers for complex workflows

{
  "mcpServers": {
    "linear": {
      "command": "npx",
      "args": ["-y", "@linear/mcp-server"],
      "env": { "LINEAR_API_KEY": "..." }
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_TOKEN": "..." }
    },
    "playwright": {
      "command": "npx",
      "args": ["-y", "@playwright/mcp-server"]
    }
  }
}

Pattern 3: Custom Resource Server

Use Case: Expose internal documentation or data

from fastmcp import FastMCP
from fastmcp.transports.stdio import serve_stdio
import asyncio
import os

mcp = FastMCP("Internal Docs", version="1.0.0")

@mcp.resource("docs://{path}")
def get_documentation(path: str) -> str:
    """Get internal documentation by path"""
    docs_dir = "/company/docs"
    filepath = os.path.join(docs_dir, path)

    if not os.path.exists(filepath):
        return f"Documentation not found: {path}"

    with open(filepath, 'r') as f:
        return f.read()

@mcp.tool()
def search_docs(query: str) -> list:
    """Search internal documentation"""
    # Implementation would use search index
    return [
        {"title": "Getting Started", "path": "onboarding/getting-started.md"},
        {"title": "API Reference", "path": "api/reference.md"}
    ]

asyncio.run(serve_stdio(mcp))

Troubleshooting MCP Integration

Issue 1: MCP Server Not Connecting

Symptoms: Tools not appearing in Claude Code

Solutions:

1. Check configuration file location
2. Verify command and args are correct
3. Test server independently: python server.py
4. Check logs in Claude Code
5. Ensure environment variables are set

Issue 2: Tool Execution Failures

Symptoms: Tool calls return errors

Solutions:

1. Validate tool parameter schemas
2. Add error handling in tool implementation
3. Check server logs for exceptions
4. Test with simple inputs first
5. Verify authentication/API keys

Issue 3: Performance Issues

Symptoms: Slow tool responses

Solutions:

1. Add caching for repeated calls
2. Implement connection pooling
3. Use async/await properly
4. Add timeout handling
5. Consider HTTP transport for remote servers

Best Practices Summary

1. Start Simple: Begin with a single tool, expand gradually
2. Use Official SDKs: Leverage maintained libraries (Python, TypeScript, C#, Java)
3. Implement Security: Validate inputs, authenticate, rate limit
4. Handle Errors Gracefully: Return structured error responses
5. Document Thoroughly: Provide clear tool descriptions and parameter schemas
6. Test Extensively: Write integration tests for all tools
7. Monitor Performance: Track response times and error rates
8. Version Your Servers: Use semantic versioning for changes
9. Leverage Context7: Research documentation before implementation
10. Follow MCP Specification: Adhere to official protocol standards

Quick Reference

MCP Server Checklist

• Server name and version defined
• Tools have clear descriptions
• Parameter schemas are complete
• Error handling implemented
• Authentication/authorization configured
• Logging enabled
• Tests written
• Documentation created
• Security validated
• Performance tested

Essential Commands

# Install MCP SDKs
pip install modelcontextprotocol
npm install @modelcontextprotocol/sdk

# Test server independently
python server.py

# Verify Claude Code config
cat ~/Library/Application\ Support/Claude/claude_desktop_config.json

# Research with Context7
/ctx7 [library-name]

Resources

Official Documentation:

• MCP Specification: https://modelcontextprotocol.io/specification
• MCP GitHub: https://github.com/modelcontextprotocol
• MCP Servers: https://github.com/modelcontextprotocol (servers directory)

SDKs:

• Python SDK: https://github.com/modelcontextprotocol/python-sdk
• TypeScript SDK: https://github.com/modelcontextprotocol/typescript-sdk
• C# SDK: https://github.com/modelcontextprotocol/csharp-sdk
• Java SDK: https://github.com/modelcontextprotocol/java-sdk

Learning Resources:

• Microsoft MCP for Beginners: https://github.com/microsoft/mcp-for-beginners
• Anthropic MCP Announcement: https://www.anthropic.com/news/model-context-protocol

Skill Version: 1.0.0 Last Updated: 2025-10-18 Context7 Research: Microsoft MCP for Beginners (Trust: 9.9)