mcpcensus

Learn

Short, evidence-first notes on MCP trust, health, and install hygiene. Built for humans and agents. Numbers always come from the live census — never invented.

First gate, not a full security audit · how scores work

What changed in MCP 2026-07-28

The stateless revision removed the initialize handshake and made server/discover mandatory. What it means if you run an MCP server.

2026-07-31

Agent checklist: install an MCP server safely

For coding agents and humans driving them — a short runbook.

2026-07-23

How to evaluate an MCP server before you install it

A practical pre-install gate that pairs census checks with human review.

2026-07-23

How to tell if an MCP server is official

Brand match is an evidence problem, not a naming contest.

2026-07-23

Look-alike MCP namespaces and brand risk

Why a verified domain is not the same as vendor affiliation.

2026-07-23

MCP registry vs The MCP Census

The registry lists servers. The census checks health and ownership with evidence.

2026-07-23

MCP server health signals that actually matter

Archived repos, deprecated packages, dead remotes — fact-based signals.

2026-07-23

MCP supply-chain checklist for platform teams

Inventory and gate MCP servers like any other plugin supply chain.

2026-07-23

MCP trust score explained

A published point sum with factors — not a vibes rating.

2026-07-23

Official vs Verified vs trust score

Three different claims on an MCP Census card — plain English.

2026-07-23

Remote vs local MCP servers

Hosted endpoints and installable packages are different risk shapes.

2026-07-23

What is an MCP server?

Plain-language definition of Model Context Protocol servers and why install choices matter.

2026-07-23

Why MCP servers need a census

Directories scale discovery. Trust needs continuous evidence.

2026-07-23

Want the live table? Open lookup. Agents: llms.txt · API.