Is that MCP server actually up?
MCP Health checks remote MCP servers from the official registry with the real client handshake, keeps their uptime and latency history, and shows real download counts for MCP packages. Find a server you can rely on, or prove that yours is one.
How it works
Sync the registry
Every server in the official MCP registry, with its remote endpoints and npm, PyPI and Docker packages, refreshed every 15 minutes.
Check like a real client
initialize, tools/list and OAuth discovery, exactly what Claude, Cursor or your agent does when it connects. No credentials, no tool calls.
Keep the history
Uptime over 24 hours, 7 and 30 days, 90-day bars, latency, incidents, an A–F spec grade, and daily downloads straight from the package registries.
The registry lists tens of thousands of servers, so the hottest categories are monitored on a schedule and everything else is checked when someone asks. Checking a server keeps it monitored for 7 days.
Use it to…
Pick MCP servers that stay up
Before you add a server to Claude, Cursor or your agent, see whether it actually answers, how fast, whether its sign-in works, and how many people install it.
Browse by uptime →Show your server is healthy
Run the checker against your endpoint, fix what the grade points out, and add a live badge to your README. It stays green while your server answers.
Wire it into your tools
Everything is JSON: search servers, read uptime and incidents, or run a check from CI. Subscribe to the Atom feed to hear when a server you depend on goes down.
GET /api/servers?q=githubAPI reference →
Most downloaded
All servers →Do things on live websites: prices, availability, quotes, bookings, anything behind a form or login.
Double-entry books an agent keeps as the user: invoices, VAT returns, 130 country packs.
The source of direction for your agents: what to build next, its spec and the calls you made.
Attestari's verdict on an npm or PyPI package version, asked before an agent installs it.
How often ChatGPT, Claude, Gemini and Perplexity name and cite your brand, with the evidence.
Agent-native web hosting — deploy sites, manage DNS, register domains, scale infrastructure
AI travel agent — book flights, hotels, activities, and events worldwide via autonomad.ai.
Affiliate network reporting in your AI client. Bring your own keys. Most adapters experimental.
Monitored right now
All monitored →run any ai model. compose agents, stack knowledge, connect tools. one api, pay per run.
Attestari's verdict on an npm or PyPI package version, asked before an agent installs it.
AI travel agent — book flights, hotels, activities, and events worldwide via autonomad.ai.
Agent-native web hosting — deploy sites, manage DNS, register domains, scale infrastructure
Do things on live websites: prices, availability, quotes, bookings, anything behind a form or login.
Double-entry books an agent keeps as the user: invoices, VAT returns, 130 country packs.
Recent incidents
All incidents →No incidents recorded yet. An incident opens when a monitored server fails 2 checks in a row.
Questions
What counts as “up”?
The endpoint answered the MCP initialize handshake, or asked for sign-in in the way the MCP auth spec describes. Anything else (timeouts, errors, pages that aren't MCP) counts as down. Uptime is the share of scheduled checks that were up.
Why isn't every server monitored?
The registry lists 876 remote endpoints, far more than can be checked every half hour. The busiest categories, by real downloads and by how often people ask about them, are monitored on a schedule. Every other server gets a check when someone requests one, and is then monitored for a week.
Where do download numbers come from?
Straight from npm, pypistats.org and Docker Hub. Registries that don't publish counts show a note instead of a number. Remote servers have no public usage numbers at all, so none are shown.
Does checking a server touch its data?
No. The checker only does what any MCP client does when it connects: it never sends credentials, guesses tokens or calls tools. Every request carries an identifying User-Agent.
Check any MCP server
Paste a remote MCP URL and get a graded report card in a few seconds: handshake, latency, tools, auth discovery and a README badge.
Open the checker