Data report · Updated weekly · Raw data (JSON)

The State of Public MCP Servers

We scan a rotating set of popular MCP servers every week with a real JSON-RPC handshake — not a ping. Here's what the current data says. All numbers below are live from this week's index.

Finding 1 — most production MCP servers require auth

Loading…

That's the correct posture for anything touching private data (Notion, Linear, Sentry, Stripe, GitHub all sit behind OAuth). But it means the openly-introspectable surface — the servers agents can discover and try anonymously — is much smaller than the raw registry counts suggest.

Finding 2 — the public servers all hit the same ceiling

Loading…

Every publicly-introspectable server we rank tops out at the same score, docked the same points for the same reason: they answer initialize without authentication. For a public docs/knowledge server that's a deliberate, reasonable choice — but it's the one thing separating them from a perfect score, and it's identical across the board.

Finding 3 — handshake latency varies wildly

Loading…

Agents chain many calls, so handshake latency compounds. The spread between the fastest and slowest public server is large — a real, measurable quality difference that never shows up in a registry listing.

Finding 4 — tool documentation is the differentiator

Across the servers that pass conformance, the thing that varies most is tool description coverage — and it's the single strongest predictor of whether agents actually call a server's tools. See why your tools aren't being called.

Method: each server gets a live initialize + tools/list handshake via MCP Pulse. Auth-gated servers can't be introspected anonymously and are excluded from scored rankings but counted in the totals. Data refreshes weekly; cite it freely with a link.

Where does your MCP server land?

Real handshake, conformance score, latency — free, 10 seconds.

Scan my server →