Skip to the page

Agno

Agno's MCPTools toolkit connects an Agno agent to tradefloor's local MCP server. That gives the agent 19 simulation and evaluation tools, such as evaluate_strategies, run_stress_scenario and explain_price_move. With the session tools a model can trade step by step in a simulated market the server keeps open, and no order reaches a real broker.

Install

pip install "agno[mcp]"

MCPTools needs fastmcp to open a connection, and the mcp extra adds it. The server runs in a separate environment through uvx, which needs uv.

Connect

MCPTools starts the server as a subprocess and registers each MCP tool as an Agno function in toolkit.functions. This example lists them, then calls list_scenarios with no model or API key.

import asyncio
import json

from agno.tools.mcp import MCPTools


async def main():
    async with MCPTools(
        command='uvx --with "tradefloor[mcp]" tradefloor mcp',
        timeout_seconds=60,
    ) as toolkit:
        names = sorted(toolkit.functions)
        print(len(names), "tools:", ", ".join(names[:4]), "...")

        result = await toolkit.functions["list_scenarios"].entrypoint()
        scenarios = json.loads(result.content)["shipped"]
        print("scenarios:", ", ".join(s["name"] for s in scenarios))


asyncio.run(main())
22 tools: build_scenario, build_universe, check_envelope, check_job ...
scenarios: curve_shock, geopolitical_conflict, liquidity_crisis, oil_price_spike, policy_regime_shift, rate_shock, recession, recession_2008

To give the tools to an agent, pass the connected toolkit inside the async with block, as in Agent(model=..., tools=[toolkit]). timeout_seconds is Agno's read timeout for the MCP client and defaults to 10. Use include_tools or exclude_tools on MCPTools to limit which tools the agent sees.

Tested with

Python 3.12.14, agno 3.1.1, fastmcp 4.0.11 and mcp 2.3.0 on macOS, and tradefloor 0.11.0.