Configuración del SDK de Agentes de OpenAI

Conecte el OpenAI Agents SDK al servidor MCP de graph8. El Agents SDK incluye soporte nativo para MCP, por lo que la conexión con graph8 solo requiere una llamada a MCPServerStreamableHttp.

Requisitos previos

Python 3.10+
SDK openai-agents con el extra de MCP

pip install 'openai-agents[mcp]'

MCP alojado en la nube (OAuth remoto)

import asyncio
import os

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


async def main() -> None:
    async with MCPServerStreamableHttp(
        name="graph8",
        params={"url": "https://be.graph8.com/mcp/"},
    ) as server:
        agent = Agent(
            name="ProspectFinder",
            instructions=(
                "You find prospects for SaaS GTM teams. Use g8_find_contacts to "
                "preview. Never call g8_build_contact_list, g8_enrich_contacts, "
                "or g8_add_to_sequence without explicit user confirmation."
            ),
            mcp_servers=[server],
            model="gpt-4o",
        )

        result = await Runner.run(
            agent,
            "Find 10 VP Engineering at Series B SaaS in the US. "
            "Return name, company, title, and linkedin URL.",
        )
        print(result.final_output)


if __name__ == "__main__":
    asyncio.run(main())

La primera ejecución solicita autenticación OAuth en el navegador. El SDK guarda el token en caché.

MCP autoalojado (stdio)

import os

from agents.mcp import MCPServerStdio


server = MCPServerStdio(
    name="graph8",
    params={
        "command": "uvx",
        "args": ["g8-mcp-server"],
        "env": {
            "G8_API_KEY": os.environ["G8_API_KEY"],
            "G8_MCP_MODE": "gtm",
        },
    },
)

Ejemplo práctico: lista de prospectos en una sola consulta

import asyncio

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


async def build_prospect_list() -> None:
    async with MCPServerStreamableHttp(
        name="graph8",
        params={"url": "https://be.graph8.com/mcp/"},
    ) as server:
        agent = Agent(
            name="GTMAgent",
            instructions=(
                "Step 1: g8_find_contacts to preview 50 VP Eng at Series B SaaS.\n"
                "Step 2: Pick the top 25 by signal score. Ask the user to confirm.\n"
                "Step 3: g8_build_contact_list named 'Series B SaaS VP Eng'.\n"
                "Step 4: g8_list_sequences to find 'New SaaS Outreach'.\n"
                "Step 5: g8_add_to_sequence after confirmation.\n"
            ),
            mcp_servers=[server],
            model="gpt-4o",
        )
        result = await Runner.run(agent, "Run the full prospect-to-sequence flow.")
        print(result.final_output)


asyncio.run(build_prospect_list())

El componente Runner del Agents SDK gestiona los ciclos de llamadas a herramientas de forma automática. La regla de confirmación de graph8 para herramientas que consumen créditos se presenta al usuario como mensajes en lenguaje natural antes de cualquier guardado.