xCLAUDE Gateway
Dev toolsDesktopAI-assisted
See what actually happens when Claude talks to your tools.
I started building xCLAUDE Gateway because the more I used Claude with MCPs, the more I wanted to know what was actually happening between Claude and the tools I had connected to it.
It started as a way to keep a local record of MCP traffic on my Mac, mostly so I could go back after a session and see what had happened. Then I kept adding things I wanted for myself: detection for credentials and PII, prompt injection, data exports, email sends and changes to tool manifests. Later I added Claude Code too, so its tool calls can be audited alongside MCP activity.
It doesn’t block anything. The idea is to keep a trail of what actually happened, not to decide what Claude should or shouldn’t be allowed to do.
Everything runs locally on the Mac, with no account or telemetry. It’s still in beta, it’s open source under MIT, and I’d genuinely like people to try it and tell me what I’ve missed.
What worked
2 votesNobody who liked it left a reason.
What didn't
3 votes- Obvious AI tells2
- Generic template2
- Soulless1
- Submitted
- Built with
- Claude code
- Source
- github.com