OpenAI Agents SDK — connect to an Agora Workbench server¶

A minimal walkthrough of just the connection plumbing between an openai-agents SDK agent and a running Agora Workbench MCP server. The focus is the wiring between the SDK and the workbench; build the agent itself by following the OpenAI Agents SDK docs.

What you'll do¶

Start the chemistry MCP server locally on port 8020.
Wrap it in an MCPServerStreamableHttp.
Hand it to whatever agents.Agent you build.

┌────────────────────────────┐    Streamable HTTP    ┌──────────────────────────┐
│  your agents.Agent         │ ────────────────────► │  chemistry MCP server    │
│  (MCPServerStreamableHttp) │   /mcp + Bearer       │  127.0.0.1:8020/mcp      │
└────────────────────────────┘                       └──────────────────────────┘

Prerequisites¶

uv installed.
Docker — the chemistry server runs as a local container.
openai-agents installed:
```
uv sync --extra openai-agents
```

Start the chemistry MCP server¶

Follow the steps in Start the chemistry MCP server to build the base image and bring the server up at http://localhost:8020/mcp.

The OpenAI Agents SDK integration point¶

Wrap the running server in an MCPServerStreamableHttp and pass it to whatever agent you build:

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

server = MCPServerStreamableHttp(
    name="chemistry",
    params={
        "url": "http://localhost:8020/mcp",
        "headers": {"Authorization": "Bearer dev-token"},
    },
    cache_tools_list=True,
)

# Build your agent however the OpenAI Agents SDK docs describe.
agent = Agent(
    name="my_agent",
    model=model,
    mcp_servers=[server],
    instructions="...",
)

async with server:
    result = await Runner.run(agent, "What does aspirin look like?")

Key options:

Argument	What it does
`params["url"]`	Streamable-HTTP MCP endpoint — always ends in `/mcp`.
`params["headers"]`	Set the `Authorization` header here.
`cache_tools_list=True`	Skip re-listing tools on every turn — recommended once you've confirmed the server's tool set.

async with server: opens the underlying MCP session. The Runner re-uses it for the entire agent run and closes it on exit.

Where to go next¶

Build the agent itself: follow the OpenAI Agents SDK docs — pick a model, write instructions, call Agent(mcp_servers=[server]). The server is exactly the MCPServerStreamableHttp you built here.
Inject the workbench skill: the workbench runtime skill is a portable system-prompt block that teaches any agent how to use Workbench tools correctly. Append it to your agent's instructions.
Authenticate against a real server: swap the dev bearer for a real token — see Authentication.

Cleanup¶

See Cleanup in the shared server guide.