Agency Swarm supports serving your tools as production-ready MCP (Model Context Protocol) endpoints.
This enables AI models to interact with your tools remotely over HTTP, making them accessible to any MCP-compatible client or AI system.

​
Setting Up MCP Server

To expose tools as a standalone MCP endpoint, use the run_mcp() function from the integrations module.
This will create an MCP server that will serve provided tools over the streamable HTTP protocol.

​
Example: Serving Individual Tools

from pydantic import Field
from agency_swarm import BaseTool
from agency_swarm.integrations import run_mcp


class TestTool(BaseTool):
    input: str = Field(..., description="The input to the tool")

    def run(self):
        return "Test tool called with input: " + self.input

# Start MCP server with individual tools
run_mcp(tools=[TestTool])

# Or provide tools folder instead
run_mcp(tools="path/to/tools/directory")

​
Configuration Options

The run_mcp() function accepts several configuration parameters:

run_mcp(
    tools=[ExampleTool, TestTool],  # List of tools or directory path
    host="0.0.0.0",                 # Host to bind server to
    port=8000,                      # Port to bind server to
    app_token_env="APP_TOKEN",      # Environment variable for auth token
    server_name="mcp-tools-server", # MCP server identifier
    cors_origins=["*"],             # CORS allowed origins
    return_app=False,               # Return app instead of running server
)

​
Authentication

Authentication is controlled via environment variables:

import os
os.environ["APP_TOKEN"] = "your-secret-token"  # Or set in .env file

If no APP_TOKEN is set, authentication will be disabled and the server will accept all requests.

​
MCP Client Usage

You can interact with your tools using any MCP-compatible client. Do not try to make direct http calls to the server, all requests should be managed by a respective client:

from agency_swarm import Agency, Agent
from agency_swarm.tools.mcp import MCPServerStreamableHttp

# Create MCP client (make sure tool server is already running)
client = MCPServerStreamableHttp(
    name="ToolsClient",
    params={
        "url": "http://localhost:8000/mcp", 
        "headers": {"Authorization": "Bearer your-secret-token"}
    },
)

# Call TestTool
response = client.call_tool("TestTool", {"input": "Hello MCP!"})
print(f"TestTool response: {response.content[0].text}")

# Or provide this client to an Agent
agent = Agent(name="test", mcp_servers=[client])
agency = Agency([agent], name="test1")
agency.run_demo()