mcp-builder

MCP Builder - Model Context Protocol Server Development

What is MCP?

Model Context Protocol (MCP) is an open standard created by Anthropic that enables AI assistants like Claude to securely connect to external data sources and tools. Think of it as a universal adapter that allows Claude to interact with any system, API, or data source through a standardized interface.

Key Benefits:

• Standardization: One protocol for all integrations
• Security: Built-in authentication and permission controls
• Flexibility: Support for tools, resources, and prompts
• Scalability: Designed for production workloads
• Modularity: Create reusable MCP servers for different domains

Architecture Overview

MCP follows a client-server architecture:

┌─────────────┐         ┌─────────────┐         ┌──────────────┐
│   Claude    │ ←──MCP──→ │ MCP Server  │ ←──────→ │ External API │
│  (Client)   │         │  (Your Code) │         │  Database    │
└─────────────┘         └─────────────┘         └──────────────┘

Components:

• Client: Claude Desktop, Claude Code, or custom applications
• Server: Your MCP implementation (Python, TypeScript, etc.)
• Transport: Communication channel (stdio, HTTP, SSE)
• Protocol: Standardized message format (JSON-RPC 2.0)

For detailed protocol specification, see [Protocol Specification Reference](./references/protocol-specification.md).

Core Components

1. Tools: Exposing Functions Claude Can Call

Tools are the primary way to give Claude new capabilities. Each tool is a function that Claude can invoke with specific arguments.

Tool Definition Structure:

{
    "name": "tool_name",
    "description": "Clear description of what this tool does",
    "inputSchema": {
        "type": "object",
        "properties": {
            "param1": {
                "type": "string",
                "description": "Description of parameter"
            }
        },
        "required": ["param1"]
    }
}

Key Principles:

• Clear naming: Use descriptive, action-oriented names (e.g., search_database, not db_query)
• Comprehensive descriptions: Explain what the tool does, when to use it, and what it returns
• Strong schemas: Use JSON Schema to validate inputs and guide Claude
• Error handling: Return clear error messages when things go wrong

For complete schema design patterns and best practices, see [Tool Schema Reference](./references/tool-schemas.md).

2. Resources: Providing Data/Documentation Access

Resources allow Claude to access files, documentation, or structured data. Unlike tools (which perform actions), resources provide information.

Resource Types:

• Static: Fixed content (e.g., documentation files)
• Dynamic: Generated on-demand (e.g., database queries)
• Templates: Parameterized resources (e.g., user profiles)

Resource URI Patterns:

file:///path/to/file.txt          # Local file
http://example.com/api/docs       # HTTP resource
custom://database/users/123       # Custom scheme
template://report/{user_id}       # Template resource

3. Prompts: Reusable Prompt Templates

Prompts are pre-defined message templates that users can invoke. They help standardize common workflows and best practices.

Prompt Definition:

{
    "name": "code_review",
    "description": "Comprehensive code review checklist",
    "arguments": [
        {
            "name": "language",
            "description": "Programming language",
            "required": True
        }
    ]
}

4. Authentication Methods

MCP supports multiple authentication methods:

• No Authentication (development only)
• API Key Authentication (simple, medium security)
• OAuth 2.0 (third-party, high security)
• Bearer Token (API-to-API, high security)

For complete security implementation guides, see [Security Best Practices](./references/security-guide.md).

Server Implementation Workflow

Phase 1: Project Setup

Create your MCP server project:

# Create project directory
mkdir my-mcp-server
cd my-mcp-server

# Initialize Python project
uv init
uv add mcp

# Create server file
touch server.py

Phase 2: Basic Server Structure

Minimal working server:

from mcp.server import Server
from mcp.server.stdio import stdio_server
from mcp.types import Tool, TextContent
import asyncio

app = Server("my-mcp-server")

@app.list_tools()
async def list_tools():
    return [
        Tool(
            name="my_tool",
            description="Description of what this tool does",
            inputSchema={
                "type": "object",
                "properties": {
                    "param": {"type": "string"}
                },
                "required": ["param"]
            }
        )
    ]

@app.call_tool()
async def call_tool(name: str, arguments: dict):
    if name == "my_tool":
        param = arguments["param"]
        result = f"Processed: {param}"
        return [TextContent(type="text", text=result)]

async def main():
    async with stdio_server() as (read_stream, write_stream):
        await app.run(read_stream, write_stream, app.create_initialization_options())

if __name__ == "__main__":
    asyncio.run(main())

Phase 3: Tool Registration and Handlers

Registration Pattern:

@app.list_tools()
async def list_tools():
    return [
        Tool(
            name="calculator_add",
            description="Add two numbers",
            inputSchema={
                "type": "object",
                "properties": {
                    "a": {"type": "number", "description": "First number"},
                    "b": {"type": "number", "description": "Second number"}
                },
                "required": ["a", "b"]
            }
        )
    ]

Handler Pattern:

@app.call_tool()
async def call_tool(name: str, arguments: dict):
    if name == "calculator_add":
        return await handle_calculator_add(arguments)
    else:
        raise ValueError(f"Unknown tool: {name}")

async def handle_calculator_add(arguments: dict):
    a = arguments["a"]
    b = arguments["b"]
    result = a + b
    return [TextContent(type="text", text=f"{a} + {b} = {result}")]

Phase 4: Resource Implementation

Static and dynamic resource examples:

from mcp.types import Resource, ResourceContents, TextResourceContents

@app.list_resources()
async def list_resources():
    return [Resource(uri="file:///docs/readme.md", name="README",
                     description="Documentation", mimeType="text/markdown")]

@app.read_resource()
async def read_resource(uri: str):
    if uri.startswith("file://"):
        with open(uri[7:], 'r') as f:
            return ResourceContents(contents=[TextResourceContents(
                uri=uri, mimeType="text/markdown", text=f.read())])

See [Resource Server Example](./examples/resource-server.md) for complete implementation.

Phase 5: Error Handling and Testing

Error Response Pattern:

async def call_tool(name: str, arguments: dict):
    try:
        return [TextContent(type="text", text=await execute_tool(name, arguments))]
    except ValueError as e:
        return [TextContent(type="text", text=f"Invalid input: {str(e)}", isError=True)]
    except Exception as e:
        logger.exception("Unexpected error")
        return [TextContent(type="text", text=f"Error: {type(e).__name__}", isError=True)]

Testing:

# Test with MCP inspector
npx @modelcontextprotocol/inspector python server.py

See [Testing and Debugging Guide](./references/testing-debugging.md) for comprehensive strategies.

Phase 6: Claude Desktop Integration

Configuration: Edit claude_desktop_config.json:

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

{
  "mcpServers": {
    "my-server": {
      "command": "python",
      "