OpenAI Responses and Agents MCP setup

Use OpenAI's remote MCP tool with the public server URL. The application supplies the authorization token.

Responses API shape

from openai import OpenAI

client = OpenAI()
response = client.responses.create(
    model="OPENAI_MODEL",
    tools=[
        {
            "type": "mcp",
            "server_label": "orkestia_workflow",
            "server_url": "https://mcp.orkestia.dev/mcp",
            "authorization": ORKESTIA_ACCESS_TOKEN,
            "require_approval": "always",
        }
    ],
    input="List workflow types for my organization",
)

Use a user access token for user-driven actions and a scoped agent token for unattended workflows.

Approval policy

Set require_approval conservatively until the tool allowlist and destructive-tool behavior are validated for your product surface.

Smoke test

1. Create a response with the MCP tool configured.
2. Approve the tool call if prompted.
3. Confirm whoami and list_workflow_types complete.
4. Confirm the approval trail is visible in OpenAI and Orkestia audit logs.

Official docs

• OpenAI remote MCP
• OpenAI Responses remote MCP reference