Skip to content
PERMISSION/PROTOCOL
← Integrations

Workflow integration guide

Choose the authorization boundary for OpenAI Agents.

Start with one consequential action. Identify the tool or service that actually executes it, then choose a controlled path that your agents must use.

Review MCP Guard setup Walk through an approval

What setup requires.

  1. 01

    Identify one tool and the credentials it uses to reach the target system.

  2. 02

    If your workflow uses an MCP adapter with a launchable stdio server, evaluate routing that server through MCP Guard.

  3. 03

    Direct function calls require a separately implemented and tested boundary. A framework name or model setting does not install enforcement.

  4. 04

    Test approval, denial, retries, changed arguments, and any alternate path before relying on the integration.

Coverage and limits.

  • This is an integration planning guide, not a claim of a bundled framework-specific adapter.
  • MCP routing does not cover native function calls, browser actions, shell access, or other services outside the configured path.
  • Confirm the deployed record format and signing behavior if you need independently verifiable human-approval evidence.

Have a different execution path?

Bring the action, target system, and required evidence. We can assess the integration work before you commit to a rollout.

Discuss your workflow