MCP health

Why your MCP server fails directory health checks — measured, not guessed.

Directories (Glama, PulseMCP, Smithery, list repos) decide whether a server is healthy by calling it. This service runs the same five probes they do and tells you exactly which one fails, with the fix.

Check a server

curl "https://mcp-health.rowu.workers.dev/api/check?url=https://your-server.example/mcp"
curl "https://mcp-health.rowu.workers.dev/report.md?url=https://your-server.example/mcp"

What it checks

Legacy handshake — health checks identify as legacy clients and omit MCP-Protocol-Version. Rejecting it is what kept our own server unhealthy for three days.
Anonymous tools/list — measured across 110 servers: every healthy one exposes tools; most unhealthy ones expose none.
tools/call stays locked — anonymous execution is a billing and abuse risk.
WWW-Authenticate on 401 — a bare 401 looks broken to crawlers.
resource_metadata (RFC 9728) — most authenticated servers advertise it.

Scan results

See the full scan · JSON

Use it from an agent

POST https://mcp-health.rowu.workers.dev/mcp
{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}
tools: check_mcp_server(url) · list_scan_results(verdict?)

Data: every verdict comes from a live HTTP probe run at request time. No estimates.