MCP Trust Registry

A security-scanned MCP server directory, not just another list

The MCP Trust Registry catalogs Model Context Protocol servers — the connectors that let AI assistants such as Claude, Cursor and other LLM clients reach real systems: databases, browsers, cloud and DevOps tooling, memory stores and web search. Most MCP directories only aggregate links and download counts. Here, every listed server has first been analyzed by the open-source mcptrustchecker engine: the published npm or PyPI package is scanned deterministically, the same input always yields the same result, and nobody can pay to be listed or ranked higher.

How to pick a trusted MCP server using three axes

Each entry exposes three signals worth reading together before you install anything:

• Trust Score — an A-F, 0-100 grade for how safe the server is for you to adopt: the threat score of its published code, adjusted down for the capability blast radius, how strongly the publisher is verified, and how much the scan could inspect. Treat it as a first filter, then open the individual findings and their evidence.
• Capability profile — the "blast radius": whether the server can touch the filesystem, spawn processes, or make network calls, and how those combine.
• Coverage — how much of the package the scanner could actually analyze, which tells you how much weight the grade deserves.

Prefer the narrowest capability set that still does the job: an assistant that only needs read-only analytics has no reason to run a server with shell access. Anything not yet listed can be checked on demand through the free scan API.

Why security matters for Model Context Protocol servers

An MCP server sits inside the trust boundary of your AI assistant, so its risks differ from ordinary dependencies. Tool poisoning hides instructions in tool descriptions that the model reads and obeys. Prompt injection arrives through fetched pages, tickets or emails and can redirect the assistant mid-task. And an over-permissioned server turns either attack into data exfiltration: a connector that can both read your files and make outbound requests can quietly ship secrets to an attacker. Scores here are automated opinions derived from published package code — evidence to weigh, not certifications.

Is a high Trust Score a guarantee that an MCP server is safe?

No. A strong grade means the deterministic scan found no dangerous patterns in the published package at the scanned version, with the stated coverage. New releases change code, so re-check after updates, review the capability profile against your own threat model, and grant each server the least access it needs.