The MCP Census

The MCP Census is a live lookup service and REST and MCP API operated by Jaco Veldsman in Australia. It checks Model Context Protocol servers against public evidence such as registry entries, GitHub, npm, PyPI and hosted endpoints, and returns a PASS, REVIEW, BLOCK or UNKNOWN verdict. The site states it is a first gate, not a malware scan, pen-test or the official MCP registry.

What It Is

The MCP Census is a pre-install check for MCP servers. A user or agent searches for a server, or submits a whole MCP config, and gets a verdict with reasons and evidence. UNKNOWN means the census has no row for that entry, and it is not a pass. The homepage says it tracks 56,966 servers, and it separates healthy servers, verified problems, unknown and not-yet-assessed ones.

How Agents Use It

An agent can call census_stamp or census_audit_config over the MCP endpoint, or POST a .mcp.json to the audit endpoint to get one verdict per server. A one-line rule in CLAUDE.md, AGENTS.md or .cursor/rules tells the agent to run a stamp before adding, enabling or calling any server, stopping on BLOCK and asking on REVIEW. A GitHub Action can fail a pull request on BLOCK.

Change Feed and Health Signals

A cursor-based change feed reports new registry entries, remote endpoints going down or returning, tool list changes, and health or verification flips. A webhook watch can push changes for one server. Verified problems are facts such as a deleted or archived repo, a deprecated package, or no push in 6+ months.

Learning Resources

The Learn section has guides on judging whether a server is official, remote versus local servers, look-alike namespaces, health signals, trust scores and supply-chain checklists. It also has comparisons with Smithery, Glama and mcp.so.