MCPサーバーのスキャフォールド

Generate a complete, runnable MCP server project from a tool specification document, using the official MCP SDK for TypeScript or Python.

使用タイミング

• You have a tool specification (from analyze-codebase-for-mcp or written manually) and need a working server
• Starting a new MCP server project and want correct structure from the start
• Migrating an existing tool integration to the MCP protocol
• Prototyping a tool surface to test with Claude Code before full implementation
• Need both the server scaffold and a test harness for CI

入力

• 必須: Tool specification document (YAML or JSON with tool names, parameters, return types)
• 必須: Target language (typescript or python)
• 必須: Transport type (stdio or sse)
• 任意: Output directory (default: current directory)
• 任意: Package name and version
• 任意: Authentication method (none, bearer-token, api-key)
• 任意: Docker packaging (true or false, default: false)

手順

ステップ1: Select SDK Language and Transport

1.1. Choose the implementation language based on project context:

• TypeScript: Best for Node.js ecosystems, web-adjacent tools, JSON-heavy workloads
• Python: Best for data science, ML, and scientific computing tool surfaces

1.2. Choose the transport mechanism:

• stdio: Default for local tool execution. Claude Code launches the server as a subprocess.
• SSE (Server-Sent Events): For remote/shared servers. Requires HTTP hosting.

1.3. Determine authentication requirements:

• none: Local stdio servers (process-level trust)
• bearer-token: Remote SSE servers with static tokens
• api-key: Remote servers with per-client keys

期待結果: Clear language, transport, and auth choices documented.

失敗時: If requirements are ambiguous, default to TypeScript + stdio + no auth for fastest time-to-working-server.

ステップ2: Initialize Project Structure

2.1. Create the project directory and initialize:

TypeScript:

mkdir -p $PROJECT_NAME && cd $PROJECT_NAME
npm init -y
npm install @modelcontextprotocol/sdk zod
npm install -D typescript @types/node tsx
npx tsc --init --target ES2022 --module nodenext --moduleResolution nodenext --outDir dist

Python:

mkdir -p $PROJECT_NAME && cd $PROJECT_NAME
python -m venv .venv
source .venv/bin/activate
pip install mcp pydantic

2.2. Create the standard directory structure:

$PROJECT_NAME/
├── src/
│   ├── index.ts|main.py      # Server entry point
│   ├── tools/                 # One file per tool category
│   │   ├── index.ts|__init__.py
│   │   └── [category].ts|.py
│   └── utils/                 # Shared utilities
│       └── validation.ts|.py
├── test/
│   ├── harness.ts|.py         # MCP test harness
│   └── tools/
│       └── [category].test.ts|.py
├── package.json|pyproject.toml
├── tsconfig.json              # TypeScript only
├── Dockerfile                 # If Docker requested
└── README.md

2.3. Add a bin entry for npm (TypeScript) or entry point for Python:

TypeScript package.json:

{
  "name": "$PACKAGE_NAME",
  "version": "1.0.0",
  "type": "module",
  "bin": { "$PACKAGE_NAME": "./dist/index.js" },
  "scripts": {
    "build": "tsc",
    "start": "node dist/index.js",
    "dev": "tsx src/index.ts",
    "test": "tsx test/harness.ts"
  }
}

期待結果: A buildable project skeleton with all dependencies installed.

失敗時: If npm/pip install fails, check network connectivity and registry access. For TypeScript, ensure Node.js >= 18. For Python, ensure Python >= 3.10.

ステップ3: Implement Tool Handlers from Spec

3.1. Parse the tool specification document and for each tool, generate a handler:

TypeScript handler template:

import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { z } from "zod";

export function registerTools(server: McpServer): void {
  server.tool(
    "tool_name",
    "Tool description from spec",
    {
      param1: z.string().describe("Parameter description"),
      param2: z.number().optional().default(10).describe("Optional param"),
    },
    async ({ param1, param2 }) => {
      try {
        // TODO: Implement tool logic
        const result = await performAction(param1, param2);
        return {
          content: [{ type: "text", text: JSON.stringify(result, null, 2) }],
        };
      } catch (error) {
        return {
          content: [{ type: "text", text: `Error: ${(error as Error).message}` }],
          isError: true,
        };
      }
    }
  );
}

Python handler template:

from mcp.server import Server
from mcp.types import Tool, TextContent
from pydantic import BaseModel

class ToolNameParams(BaseModel):
    param1: str
    param2: int = 10

async def handle_tool_name(params: ToolNameParams) -> list[TextContent]:
    try:
        result = await perform_action(params.param1, params.param2)
        return [TextContent(type="text", text=json.dumps(result, indent=2))]
    except Exception as e:
        return [TextContent(type="text", text=f"Error: {e}")]

3.2. Generate one handler file per tool category from the specification.

3.3. Add input validation beyond type checking:

• String length limits
• Numeric range bounds
• Enum value constraints
• Required field enforcement

3.4. Add structured error responses for all anticipated failure modes.

期待結果: A handler file per category with typed parameters and error handling.

失敗時: If the spec contains ambiguous types, default to string and add a TODO comment for manual refinement.

ステップ4: Configure Transport

4.1. Create the server entry point with the chosen transport:

stdio (TypeScript):

import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import { registerTools } from "./tools/index.js";

const server = new McpServer({
  name: "$PACKAGE_NAME",
  version: "1.0.0",
});

registerTools(server);

const transport = new StdioServerTransport();
await server.connect(transport);

SSE (TypeScript):

import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { SSEServerTransport } from "@modelcontextprotocol/sdk/server/sse.js";
import { registerTools } from "./tools/index.js";

const server = new McpServer({
  name: "$PACKAGE_NAME",
  version: "1.0.0",
});

registerTools(server);

const transport = new SSEServerTransport("/messages", response);
await server.connect(transport);

4.2. If authentication is required, add middleware:

• Bearer token: validate Authorization header
• API key: validate X-API-Key header

4.3. Add a shebang line for stdio servers to enable direct execution:

#!/usr/bin/env node

期待結果: A working entry point that starts the MCP server on the configured transport.

失敗時: If the SDK version does not match the import paths, check the @modelcontextprotocol/sdk version and adjust imports. The SDK restructured paths between versions.

ステップ5: Create Test Harness

5.1. Build a test harness that validates every tool:

import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { InMemoryTransport } from "@modelcontextprotocol/sdk/inMemory.js";
import { Client } from "@modelcontextprotocol/sdk/client/index.js";

async function runTests(): Promise<void> {
  const server = createServer();
  const [clientTransport, serverTransport] = InMemoryTransport.createLinkedPair();

  await server.connect(serverTransport);
  const client = new Client({ name: "test-client", version: "1.0.0" });
  await client.connect(clientTransport);

  // Test: tools/list returns all expected tools
  const tools = await client.listTools();
  console.assert(tools.tools.length === EXPECTED_TOOL_COUNT);

  // Test: each tool with valid input
  for (const tool of tools.tools) {
    const result = await client.callTool({
      name: tool.name,
      arguments: getTestInput(tool.name),
    });
    console.assert(!result.isError, `${tool.name} failed`);
  }

  // Test: each tool with invalid input returns isError
  for (const tool of tools.tools) {
    const result = await client.callTool({
      name: tool.name,
      arguments: getInvalidInput(tool.name),
    });
    console.assert(result.isError, `${tool.name} should reject invalid input`);
  }

  console.log("All tests passed");
}

5.2. Create test fixtures for each tool: valid inputs, invalid inputs, and edge cases.

5.3. Add a test script to package.json or pyproject.toml.

期待結果: A test harness that exercises every tool with both valid and invalid inputs.

失敗時: If InMemoryTransport is not available in the SDK version, fall back to spawning the server as a subprocess and communicating via stdio pipes.

ステップ6: Generate Documentation and Configuration

6.1. Generate a README.md with:

• Project description
• Installation instructions
• Claude Code configuration command
• Claude Desktop JSON configuration snippet
• Tool listing with descriptions and parameter schemas
• Development and testing instructions

6.2. Generate Claude Code registration command:

# stdio transport
claude mcp add $PACKAGE_NAME stdio "node" "dist/index.js"

# SSE transport
claude mcp add $PACKAGE_NAME -e API_KEY=your_key -- mcp-remote http://localhost:3000/mcp

6.3. Generate Claude Desktop configuration snippet:

{
  "mcpServers": {
    "$PACKAGE_NAME": {
      "command": "node",
      "args": ["path/to/dist/index.js"]
    }
  }
}

6.4. If Docker was requested, generate a Dockerfile:

FROM node:20-slim AS build
WORKDIR /app
COPY package*.json ./
RUN npm ci
COPY . .
RUN npm run build

FROM node:20-slim
WORKDIR /app
COPY --from=build /app/dist ./dist
COPY --from=build /app/node_modules ./node_modules
COPY --from=build /app/package.json .
ENTRYPOINT ["node", "dist/index.js"]

期待結果: Complete documentation and configuration files for immediate use.

失敗時: If the generated README has placeholder values, search the project for actual values to substitute. If Docker build fails, verify the base image matches the Node.js/Python version used.

バリデーション

• Project builds without errors (npm run build or equivalent)
• Server starts and responds to tools/list JSON-RPC request
• Every tool from the specification is registered and discoverable
• Test harness passes for all tools with valid inputs
• Test harness confirms error responses for invalid inputs
• Claude Code can connect via claude mcp add command
• README includes working installation and configuration instructions
• All generated code passes linting (if configured)

よくある落とし穴

• SDK import path changes: The @modelcontextprotocol/sdk package restructured its exports between versions. Always check the installed version's actual export paths.
• Forgetting the shebang: stdio servers invoked directly need #!/usr/bin/env node as the first line to be executable.
• Blocking the event loop: Tool handlers in TypeScript must be async. Synchronous operations block all other tool calls on the server.
• Missing type: "module" in package.json: The MCP SDK uses ESM imports. Without "type": "module", Node.js treats files as CommonJS and imports fail.
• Zod schema drift: If the tool spec evolves but Zod schemas are not updated, validation mismatches cause silent failures. Generate schemas from a single source of truth.
• stdout pollution: stdio transport uses stdout for JSON-RPC. Any console.log in tool handlers corrupts the protocol stream. Use console.error or a file logger instead.

関連スキル

• analyze-codebase-for-mcp - generate the tool specification this skill consumes
• build-custom-mcp-server - manual server implementation for complex cases
• configure-mcp-server - connect the scaffolded server to Claude Code/Desktop
• troubleshoot-mcp-connection - debug connectivity issues after deployment
• containerize-mcp-server - package the server in Docker for distribution