Peer Cash MCP Agent - Prepare custody-separated cash-outs

This example connects Agno to Peer Cash through its published MCP server.

This example connects Agno to Peer Cash through its published MCP server. The server discovers supported payout rails, reads live market-rate estimates, prepares unsigned Base USDC cash-out transactions, and tracks existing orders. It never accepts private keys, signs transactions, or broadcasts them.

peer_cash.py

"""Peer Cash MCP Agent - Prepare custody-separated cash-outs

This example connects Agno to Peer Cash through its published MCP server. The
server discovers supported payout rails, reads live market-rate estimates,
prepares unsigned Base USDC cash-out transactions, and tracks existing orders.
It never accepts private keys, signs transactions, or broadcasts them.

Prerequisites:
- Node.js 22 or newer, including ``npx``
- ``uv pip install "agno[mcp]" openai``
- ``OPENAI_API_KEY`` in the environment

Peer Cash MCP: https://github.com/zkp2p/peer-cash-mcp
"""

import asyncio
import os

from agno.agent import Agent
from agno.models.openai import OpenAIResponses
from agno.tools.mcp import MCPTools
from pydantic import BaseModel


class CashoutPlan(BaseModel):
    summary: str
    actions: list[str]
    signing_required: bool
    safety_note: str


async def run_agent(message: str) -> None:
    npx_command = "npx.cmd" if os.name == "nt" else "npx"

    async with MCPTools(
        command=f"{npx_command} -y peer-cash-mcp@0.1.2",
        timeout_seconds=60,
    ) as peer_cash:
        agent = Agent(
            name="Peer Cash Agent",
            model=OpenAIResponses(id="gpt-5.5"),
            tools=[peer_cash],
            output_schema=CashoutPlan,
            # Keep the final response structured without forcing OpenAI's strict
            # schema rules onto externally defined MCP tool schemas.
            use_json_mode=True,
            instructions=[
                "Use Peer Cash to help users move Base USDC to supported fiat payout rails.",
                "Read capabilities before assuming a platform, currency, amount, or payee format is supported.",
                "Treat estimates as approximate oracle readings, never locked quotes.",
                "For mutations, return the unsigned transaction plan for the user's wallet to inspect and sign.",
                "Never ask for a private key and never claim an unsigned transaction was submitted.",
                "After an externally signed create-deposit transaction confirms, finalize its receipt before reporting an order ID.",
            ],
            markdown=True,
        )
        await agent.aprint_response(message, stream=True)


if __name__ == "__main__":
    asyncio.run(
        run_agent(
            "Show the supported cash-out platforms and estimate the USD payout for 1000 Base USDC. Do not create an order."
        )
    )

Run the Example

1. Set up your virtual environment

   Mac

   uv venv --python 3.12
   source .venv/bin/activate
   
2. Install dependencies

   uv pip install -U "agno[mcp]" openai
   
3. Prepare Node.js

   The MCP server runs with npx. Install Node.js, then verify the commands:

   node --version
   npx --version
   
4. Export your OpenAI API key

   Mac/Linux

   export OPENAI_API_KEY="your_openai_api_key_here"
   
5. Run the example

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

   python peer_cash.py
   

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