Serverdex Data Report
The real state of live MCP servers
Most directories list MCP servers they never check. Serverdex handshakes every one with a remote endpoint about once a day. Of the 21,897 we reach, 90% answer a credential-free handshake, more than 1 in 4 ask you to sign in, and the median loads about 4.7k tokens of tool definitions before your first prompt. Here is the measured reality.
Serverdex, “The real state of live MCP servers”: 90% of 21,897 probed remote MCP servers answer a handshake, 28% of those need a sign-in (Nov 22, 2026, 15:00 UTC). https://serverdex.io/articles/state-of-live-mcp-servers- 1. Why this is the only number that is measured
- 2. 9 in 10 remote servers answer
- 3. More than 1 in 4 ask for your keys
- 4. The median server costs you ~4,800 tokens
- 5. Most run the current protocol
- 6. How to read a server before you connect it
- 7. Methodology & data
1. Why this is the only number that is measured
Every large MCP directory shows a server count. Almost none of them tell you whether those servers are alive.
A listing is a claim. A server that was published a year ago and has been down for months still sits in most catalogues with a green card and an install button. What a developer actually needs to know — does it answer, does it want my keys, how much of my context does it eat — is not on the page.
Serverdex calls every server that exposes a remote endpoint with a credential-free initialize + tools/list, about once a day, and records what came back: whether it answered, whether it asked for a sign-in, the shape and token size of its tool list, and the protocol version it speaks. Everything below is that record, read live. (Servers that only ship as a local package — an npm or PyPI install with no hosted endpoint — can't be handshaked from the outside, so the liveness figures are about the servers that run remotely.)
2. 9 in 10 remote servers answer
Of the 21,897 remote servers reached, 90% answered the handshake — 64% with no sign-in at all, 26% after one. Only 10% were unreachable or answered with something that is not MCP. The ecosystem is far more alive than the “graveyard of dead servers” framing suggests — but that last tenth is exactly what a directory should hide, and most do not.
See the servers answering right now on /remote, or run the check on any single server with the MCP health check.
3. More than 1 in 4 ask for your keys
28% of the servers that answer — 5,621 of them — require a sign-in (an OAuth flow or an API key) before they expose a single tool. That is not a red flag by itself; it is what an auth or a deploy tool does. But “needs my credentials” is the first thing worth knowing about a server, and almost nobody checks it before connecting.
The real risk with an MCP server was never a hidden instruction in a description — we scanned 309,923 tool definitions for that and found it is effectively nil (the scan is here). The risk is capability: a tool with real reach that you wired into your agent without looking. Whether it needs your keys is the first and cheapest thing to vet.
4. The median server costs you ~4,800 tokens
Every connected server's tool definitions are loaded into the model's context before it reads your prompt. That is a cost on every turn, and it is wildly uneven.
The heaviest single server loads more than a whole 200K context window of definitions before you ask it anything. The full distribution, and how to keep an agent lean, is in The MCP context tax.
5. Most run the current protocol
About 78% of answering servers speak the current 2025-06-18 MCP spec; the rest trail on older revisions. A long tail on year-old specs is the quiet maintenance signal — a server frozen on an old protocol is usually a server nobody is tending.
MCP protocol revision reported at handshake · count of answering servers
6. How to read a server before you connect it
The numbers above are the four questions worth asking of any single server, in order:
1. Does it answer? A listing is not a liveness check. Confirm it responds before you build on it — the live list and the health check do this for you.
2. Does it want your credentials? Know whether a sign-in is required, and to whom you are handing the keys.
3. What does it cost your context? A 200-tool server is a tax on every turn. Prefer lean, well-scoped ones — see the context tax.
4. Can you read its tools? The tool names and descriptions are the whole contract. If you cannot tell what a tool does, that is the finding.
7. Methodology & data
How the numbers above were produced, so you can weigh them.
Serverdex calls every listed server that exposes a remote endpoint with a credential-free initialize + tools/list, about once a day, and records the result. A server counts as answering when it completes the handshake (with or without a required sign-in); “no MCP answer” means it was unreachable or replied with something that is not the MCP protocol. Token sizes are estimated as the characters of the raw tool definitions divided by four; they measure the definitions a client sends, not runtime behaviour. Local-only packages with no hosted endpoint are not handshaked, so the liveness figures describe remote servers.
Figures on this page are read live from the directory's precomputed stats and are as of Nov 22, 2026, 15:00 UTC; 42,624 servers are listed, 6,969 added in the last 7 days; the median answering server replies in about 118 ms.
Serverdex, “The real state of live MCP servers” (November 21, 2026). https://serverdex.io/articles/state-of-live-mcp-serversFound an error, or want the per-server breakdown? Get in touch.