OpenAI Agents SDK verifiable receipts

Connect with native OpenAI Agents SDK MCP support

Use the SDK's Streamable HTTP MCP client first. The model-provider credential, if needed by your agent, is separate from Zambo; the Zambo endpoint itself needs no API key.

pip install openai-agents

import asyncio
from agents import Agent, Runner
from agents.mcp import MCPServerStreamableHttp

async def main():
    async with MCPServerStreamableHttp(
        name="zambo",
        params={"url": "https://zambo.dev/mcp"},
    ) as zambo:
        agent = Agent(name="worker", mcp_servers=[zambo])
        result = await Runner.run(
            agent,
            "Get the ETH price and return the Zambo receipt URL.",
        )
        print(result.final_output)

asyncio.run(main())

Install the optional adapter package

pip install openai-agents-zambo openai-agents

Published package: openai-agents-zambo on PyPI.

Wire the integration

ZamboRunHooks subclasses the OpenAI Agents SDK RunHooks base and records tool_start and tool_end events through the live Zambo endpoint. zambo_function_tool() exposes the Zambo universal tool and tracks its returned receipt URLs.

import asyncio
from agents import Agent, Runner
from openai_agents_zambo import ZamboRunHooks, zambo_function_tool

hooks = ZamboRunHooks()
agent = Agent(
    name="researcher",
    instructions="Use ask_zambo for the requested task.",
    tools=[zambo_function_tool()],
)

result = asyncio.run(
    Runner.run(agent, "What is the capital of France?", hooks=hooks)
)
print(result.final_output)
print("Receipts:", hooks.receipt_urls)
print("Errors:", hooks.receipt_errors)

Want receipts for your own AI work?