Multiple Servers

Use multiple MCP servers in a single agent.

The source excerpt uses the archived Brave reference server and closes connections only after successful responses. Apply the maintained-server and lifecycle changes below.

multiple_servers.py
"""
This example demonstrates how to use multiple MCP servers in a single agent.

Each server gets its own MCPTools instance; pass them all to the agent.

Prerequisites:
- Set the environment variable "BRAVE_API_KEY" for the Brave search MCP tools.
- You can get the API key from the Brave website: https://brave.com/search/api/
"""

import asyncio
from os import getenv

from agno.agent import Agent
from agno.tools.mcp import MCPTools

# ---------------------------------------------------------------------------
# Create Agent
# ---------------------------------------------------------------------------


async def run_agent(message: str) -> None:
    # Initialize one MCPTools instance per server
    airbnb_tools = MCPTools(
        command="npx -y @openbnb/mcp-server-airbnb --ignore-robots-txt",
        timeout_seconds=30,
    )
    search_tools = MCPTools(
        command="npx -y @modelcontextprotocol/server-brave-search",
        env={"BRAVE_API_KEY": getenv("BRAVE_API_KEY")},
        timeout_seconds=30,
    )

    # Connect to the MCP servers
    await airbnb_tools.connect()
    await search_tools.connect()

    # Use the MCP tools with an Agent
    agent = Agent(
        tools=[airbnb_tools, search_tools],
        markdown=True,
    )
    await agent.aprint_response(message)

    # Close the MCP connections
    await airbnb_tools.close()
    await search_tools.close()


# Example usage
# ---------------------------------------------------------------------------
# Run Agent
# ---------------------------------------------------------------------------

if __name__ == "__main__":
    asyncio.run(run_agent("What listings are available in Barcelona tonight?"))
    asyncio.run(run_agent("What's the fastest way to get to Barcelona from London?"))

Run the Example

Set up your virtual environment

uv venv --python 3.12
source .venv/bin/activate

Install dependencies

uv pip install -U "agno[mcp]" openai

Prepare Node.js

The maintained Brave server requires Node.js 22 or later. Install it, then verify node --version reports v22 or later and npx --version succeeds.

Export your API keys

export BRAVE_API_KEY="your_brave_api_key_here"
export OPENAI_API_KEY="your_openai_api_key_here"

Use the maintained Brave server

Replace npx -y @modelcontextprotocol/server-brave-search with npx -y @brave/brave-search-mcp-server --transport stdio in the saved file. Keep the existing BRAVE_API_KEY environment mapping. See Brave's maintained server.

Close connections on failures

In the saved file, replace run_agent with this version. Keep the existing imports and main block. The async context managers reject failed connections and close every entered connection if the model or a later connection fails.

async def run_agent(message: str) -> None:
    # Initialize one MCPTools instance per server
    airbnb_tools = MCPTools(
        command="npx -y @openbnb/mcp-server-airbnb --ignore-robots-txt",
        timeout_seconds=30,
    )
    search_tools = MCPTools(
        command="npx -y @brave/brave-search-mcp-server --transport stdio",
        env={"BRAVE_API_KEY": getenv("BRAVE_API_KEY")},
        timeout_seconds=30,
    )

    async with airbnb_tools, search_tools:
        agent = Agent(
            tools=[airbnb_tools, search_tools],
            markdown=True,
        )
        await agent.aprint_response(message)

Run the example

Save the code above as multiple_servers.py, then run:

python multiple_servers.py

Full source: cookbook/91_tools/mcp/multiple_servers.py