# Gage MCP server

> Enables secure, signed handoffs between AI agents using a reading frame that prevents prompt injection and verifies sender identity.

- Repository: https://github.com/warewithall2/gage
- Page: https://hostedmcp.org/servers/warewithall2/gage
- Transport: stdio
- Stars: 1
- License: Apache-2.0
- Latest version: 0.3.6

## FAQ

### How does gage prevent prompt injection?

It uses a reading frame with nonced fence markers that the sender cannot predict, ensuring the payload cannot prematurely close the frame and masquerade as the host.

### Can I use gage with my existing AI agents via MCP?

Yes, gage provides an MCP server that allows agents to call gage as tools.

### Does gage verify if a signature is valid or if I trust the person?

It does both, but distinguishes between them; a valid signature only proves authenticity, while gage computes a trust verdict based on your local allowlist and human confirmation.

### What happens if a sender is not on my allowlist?

The contents are still shown but with a verdict displayed above them, allowing for first contact with new people without being vulnerable to unverified authority.

### Is gage a transport protocol?

No, it is not a transport or a registry; it uses signed directories that can move over any shared folder, git, or email.
