Checked, not claimed.
Does it work with your AI?
Find out if a tool has an official MCP server or Claude connector, what your AI can actually do with it, and whether it’s safe to connect.
125 tools checked · 107 official · 18 community, bridge or unverified
Which AI assistants can use MCP servers?
Where each assistant takes custom MCP servers and connectors, on which plans, and how to add one.
- ClaudeTakes custom MCP4 products · by Anthropic
- ChatGPTTakes custom MCP4 products · by OpenAI
- GeminiTakes custom MCP4 products · by Google
- GrokTakes custom MCP3 products · by xAI
- MuseNot verified3 products · by Meta
4 of 5 take your own MCP servers today · checked Oct 2026
Most searched
The tools people ask about most, with what their MCP server can actually do.
Playwright
Playwright has an official MCP server published by Microsoft (microsoft/playwright-mcp, npm @playwright/mcp). It runs locally on your machine over stdio via npx, needs no account or API key, and drives a real browser. Caveat: the project states it is not a security boundary, and it has no server-wide read-only mode.
Figma
GitHub
Blender
Chrome DevTools
Latest verdict changes
Every verdict is dated and re-checked. When a status, connector or verdict changes, it shows up here first.
What the verdicts mean
Four words, used the same way on every page. Every page shows the date it was checked.
- 107tools
Official
The vendor publishes it, confirmed in their own docs or repo.
Start with the vendor’s own setup
- 18tools
Community only
Only third-party servers exist. Useful, but nobody at the vendor vouches for them.
Read the safety notes first
Via a bridge only
Reachable only through a bridge like Zapier or Composio.
Works through a middleman
Not verified
We couldn’t confirm either way. It never means “no”.
Know more? Tell us
Five checks before you connect
Every page also runs a five-point safety check: publisher, sign-in method, write access, hidden instructions in tool descriptions, and repo upkeep.
How we checkCheck 1
Publisher
Is the publisher verified? We check that the vendor owns the repo or endpoint.
Check 2
Sign-in method
How it signs in: OAuth, or an API key or token. Long-lived, broad tokens get flagged.
Check 3
Write access
Whether it can write or delete by default, or offers a read-only mode.
Check 4
Hidden instructions
Whether its tool descriptions contain instructions aimed at the AI.
Check 5
Repo upkeep
Whether the repository is maintained. Archived, or no push in six months, gets flagged.
New to MCP? Start here
MCP is an open standard that lets AI assistants like Claude use outside tools and data. Learn what a server is, what "official" means, and the five checks to run before you connect one.
Read the guides- 01What is MCP? The Model Context Protocol in plain EnglishMCP is an open standard that lets AI assistants like Claude use outside tools and data. What it is, how a server works, and what "official" means.Read guide
- 02Is MCP safe? Five checks before you connect a serverThe five safety checks MCP or Not runs on every server: who publishes it, how it signs in, what it can write or delete, hidden instructions, and upkeep.Read guide
Missing a tool? Ask us to check it
Tell us which tool to check next, or that a verdict is out of date. We check every claim against the vendor’s own docs or repository before a verdict changes. Submissions open soon: until then the form says so and stores nothing.
New official MCP servers, weekly
One short email when vendors ship, change or retire an official server or Claude connector. Checked against primary sources, like every verdict here.