Who it is for
Developers building AI agents and integrators maintaining MCP connections for their clients.
MCP is a protocol that connects an AI assistant to external tools such as search, code and business applications.
What the prototype does
The local command-line prototype inspects explicitly selected HTTPS MCP servers without authentication. It records catalogues, compares declared tools, descriptions and schemas with an approved reference, and exports a local report with evidence identifiers.
It does not execute MCP tools. Approval of a new reference remains an explicit human decision.
Current stage
Local prototype. A separate demonstration interface uses fictitious connections and simulated decisions. It is not connected to the collector.
No private hosted workspace or continuous monitoring service is available from this page.
Try the presentation demo
The interface is in French. It explains the project, shows example metadata changes and lets you simulate a decision. No real server is inspected and decisions disappear when you reload the page.
Open the interactive demoTrust Verify: check a proposed fix
For developers and technical leads reviewing a vulnerability report or a proposed fix. The demonstration compares forbidden and legitimate access before and after correction.
The interface is in French. It recalculates results in your browser using embedded fictitious functions. It does not scan websites or accept external code.
Open the Trust Verify demonstrationWhat it cannot establish
A declaration can differ from actual behaviour. Metadata inspection does not certify server security, prevent data leakage, manage agent permissions or establish where data is processed.
An unavailable observation is inconclusive. It must not be interpreted as proof that an unchanged server is safe or that a server is dangerous.
Discuss a pilot
Use the contact form to describe your tools and the changes you need to follow.
Open contact form