Free MCP Server Health Checker
AI agents reach your tools through the Model Context Protocol. Paste your MCP endpoint and we run a real handshake — is it live, spec-compliant, and what does it expose? Instantly, with no signup. Then let Merlonix watch it for you.
We open a real MCP session — a JSON-RPC initialize handshake, then tools/list — and report whether your server is live, spec-compliant, and what it exposes to agents. No signup, nothing stored.
What does the MCP health checker actually do?
It opens a real Model Context Protocol session with your endpoint — not just an HTTP ping. It sends a JSON-RPC "initialize" handshake over the Streamable HTTP transport, reads back the protocol version, server name/version, and advertised capabilities, then calls tools/list (and resources/list and prompts/list when advertised) to inventory exactly what an AI agent can call. It reports up / degraded / down, the handshake latency, and per-signal findings. An endpoint can return HTTP 200 to a browser and still be completely broken for an agent — so we speak the protocol instead of checking a status code.
Why does MCP server uptime need its own monitor?
MCP servers are how AI agents (Claude, Cursor, ChatGPT, and custom agents) call your tools and read your data. If your MCP server is down, rejects the handshake, gets slow, or silently changes a tool, every agent workflow built on it breaks — and a generic HTTP uptime check won't notice, because the URL still returns 200. A protocol-aware monitor is the only thing that catches an initialize error, an empty tools/list, or a tool-schema change. No other uptime tool checks MCP servers at all yet.
Do you check whether my MCP server is safe to hand to an AI agent?
Yes — every check now includes an A–F MCP security posture grade. It scans each tool description your server advertises for tool-poisoning signals (hidden "ignore previous instructions" directives, <important>-style instruction markup, exfiltration hints like reading .env or SSH keys, and invisible/zero-width characters), because an agent reads a tool description verbatim and a hidden instruction there can hijack it. It also grades whether the server requires authentication — roughly a quarter of public MCP servers require none — and whether it is served over HTTPS so an agent's requests and any token stay encrypted. A single poisoning flag caps the grade, and the findings tell you exactly which tool to fix. It is a deterministic heuristic, not a guarantee, and it runs on the free check with nothing stored.
What is "tool drift" and why should I care?
Drift is when your MCP server's contract changes between checks: a tool disappears, a new tool appears, a tool's input schema changes, the protocol version shifts, or your server version bumps. A silent schema change is the worst kind — the server stays "up" but every agent calling that tool starts failing, because the arguments it sends no longer match. The continuous Merlonix monitor fingerprints every tool on each run and alerts you the moment the contract changes, so you catch a breaking change before your users do.
Are you ready for the July 2026 stateless MCP spec?
Yes. The 2026-07-28 MCP specification makes the protocol stateless — it removes the "initialize" handshake (SEP-2575) and the Mcp-Session-Id header (SEP-2567), with client info and protocol version now carried per request. A server that migrates keeps working perfectly for agents but no longer completes the classic handshake, so a naive checker would report it "down". Our checker probes both: it tries the stateful handshake, and if that no longer applies it calls tools/list directly on the stateless core — then labels which spec generation your server answered on. If your server crosses from stateful to stateless between checks, the continuous monitor flags that migration as drift, because clients pinned to the old transport can break the moment it happens.
My MCP server requires authentication — can you still check it?
Yes. If your endpoint responds with 401/403, the free checker reports it as "up but auth-gated" rather than down — the server is alive, it just needs a credential. The continuous monitor lets you store an access token so it can complete the full handshake and tools/list behind your auth for a real health check, without ever exposing the token in a URL.
How do I monitor my MCP server continuously?
A one-time check tells you where you stand right now. Merlonix re-runs this exact handshake on a schedule and alerts you when your server stops responding, the handshake starts failing, latency crosses your threshold, or a tool/resource/prompt is added, removed, or changes schema — alongside SSL, DNS, domain, and uptime monitoring. Run a check above, then claim the free plan ($0, no credit card) to keep one server monitored.
Related reading
Monitor your MCP server continuously
A one-time check tells you where you stand right now. Merlonix re-runs this exact handshake on a schedule and alerts you when your server stops responding, the handshake starts failing, latency crosses your threshold, or a tool is added, removed, or silently changes schema.
No credit card required.