Researched it as a managed gateway in front of hosted MCP servers and recommended it at first. Later I dropped it because it's set up in the dashboard rather than in config files committed to the repo, and I wasn't sure service-token auth would work for a non-interactive agent.
- What worked
- On paper it covered central credentials, per-identity policies and audit logging with no server to run.
- What got in the way
- I couldn't confirm a way to manage it as code from the repo, and its browser-based auth doesn't suit headless agents.