Phase 1b: Rename all ~94 commands across 12 plugins to /<noun> <action> sub-command pattern. Git-flow consolidated from 8→5 commands (commit variants absorbed into --push/--merge/--sync flags). Dispatch files, name: frontmatter, and cross-reference updates for all plugins. Phase 2: Design documents for 8 new plugins in docs/designs/. Phase 3: Scaffold 8 new plugins — saas-api-platform, saas-db-migrate, saas-react-platform, saas-test-pilot, data-seed, ops-release-manager, ops-deploy-pipeline, debug-mcp. Each with plugin.json, commands, agents, skills, README, and claude-md-integration. Marketplace grows from 12→20. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2.8 KiB
2.8 KiB
MCP Server Patterns Skill
Standard patterns for building Python MCP servers compatible with Claude Code.
Directory Structure
mcp-servers/<name>/
├── mcp_server/
│ ├── __init__.py # Package marker
│ ├── config.py # Configuration loader
│ ├── server.py # Entry point (MCP server setup)
│ └── tools/
│ ├── __init__.py # Tool registration
│ └── <category>.py # Tool implementations grouped by domain
├── tests/
│ ├── __init__.py
│ └── test_tools.py
├── requirements.txt
└── README.md
FastMCP Pattern (Recommended)
The FastMCP decorator pattern provides the simplest server implementation:
# server.py
from mcp.server.fastmcp import FastMCP
from .config import load_config
mcp = FastMCP("server-name")
config = load_config()
# Import tools to register them
from .tools import category # noqa: F401
if __name__ == "__main__":
mcp.run()
# tools/category.py
from ..server import mcp, config
@mcp.tool()
def my_tool(param: str) -> str:
"""Tool description."""
return f"Result for {param}"
Configuration Loader Pattern
# config.py
import os
from pathlib import Path
from dotenv import load_dotenv
def load_config() -> dict:
# System-level config
sys_config = Path.home() / ".config" / "claude" / "<name>.env"
if sys_config.exists():
load_dotenv(sys_config)
# Project-level overrides
project_env = Path.cwd() / ".env"
if project_env.exists():
load_dotenv(project_env, override=True)
config = {
"api_url": os.getenv("<NAME>_API_URL"),
"api_token": os.getenv("<NAME>_API_TOKEN"),
}
# Validate required
missing = [k for k, v in config.items() if v is None]
if missing:
raise ValueError(f"Missing config: {', '.join(missing)}")
return config
Entry Point Configuration
In .mcp.json:
{
"mcpServers": {
"<name>": {
"command": "mcp-servers/<name>/.venv/bin/python",
"args": ["-m", "mcp_server.server"],
"cwd": "mcp-servers/<name>"
}
}
}
The -m flag runs the module as a script, using __main__.py or the if __name__ == "__main__" block.
Requirements
Minimum dependencies for an MCP server:
mcp>=1.0.0
python-dotenv>=1.0.0
For HTTP-based integrations add:
httpx>=0.24.0
Testing Pattern
# tests/test_tools.py
import pytest
from mcp_server.tools.category import my_tool
def test_my_tool():
result = my_tool("test")
assert "test" in result
Startup Logging
Always log initialization status to stderr:
import sys
def log(msg):
print(msg, file=sys.stderr)
log(f"MCP Server '{name}' initialized: {len(tools)} tools registered")