How to build a multi-tool MCP server
Quick answer
Use the official
mcp Python SDK to create a multi-tool MCP server by registering multiple tools and exposing them via the Server class. The server can be run with stdio_server for standard I/O communication, enabling AI agents to access various tools seamlessly.PREREQUISITES
Python 3.8+pip install mcpBasic knowledge of Python async programming
Setup
Install the official mcp Python SDK and prepare your environment to build the MCP server.
pip install mcpStep by step
Create a multi-tool MCP server by defining multiple tool handlers and registering them with the Server class. Use stdio_server to run the server over standard input/output, which is the common transport for MCP servers.
import asyncio
from mcp.server import Server
from mcp.server.stdio import stdio_server
# Define async tool handlers
async def tool_one(params):
# Example tool logic
return {"result": f"Tool One received: {params.get('input', '')}"}
async def tool_two(params):
# Another tool logic
return {"result": f"Tool Two processed: {params.get('input', '')}"}
async def main():
# Create MCP server instance
server = Server()
# Register multiple tools
server.register_tool("tool_one", tool_one)
server.register_tool("tool_two", tool_two)
# Run the server using stdio transport
await stdio_server(server)
if __name__ == "__main__":
asyncio.run(main())Common variations
You can run the MCP server with different transports like SSE or TCP, or implement synchronous tool handlers if needed. Also, you can extend the server to handle authentication or logging.
import asyncio
from mcp.server import Server
from mcp.server.stdio import stdio_server
# Synchronous tool example
def sync_tool(params):
return {"result": f"Sync tool got: {params.get('input', '')}"}
async def main():
server = Server()
# Register sync tool (wrap with async)
server.register_tool("sync_tool", lambda params: asyncio.to_thread(sync_tool, params))
await stdio_server(server)
if __name__ == "__main__":
asyncio.run(main())Troubleshooting
- If the server does not respond, ensure you run it with proper stdio pipes or use a compatible client.
- For import errors, verify
mcpis installed in your Python environment. - If tools do not register, confirm you call
register_toolbefore starting the server.
Key Takeaways
- Use the official
mcpPython SDK to build multi-tool MCP servers efficiently. - Register each tool with
Server.register_toolbefore running the server. - Run the server with
stdio_serverfor standard I/O communication. - Async tool handlers are preferred but synchronous tools can be wrapped with
asyncio.to_thread. - Troubleshoot by verifying environment setup and transport compatibility.