MCP server observations, late October 2026
This is the second issue of our monthly series. Each month we publish the same handful of numbers about the public MCP servers we observe, read from our production database on a stated date, with the definitions at the end. The first issue covered 1 October to 21 October 2026, 20:00 UTC. This one picks up where it stopped and runs to the end of the month: 21 October, 20:00 UTC, to 1 November 2026, 00:00 UTC.
The list
On 31 October our list held 5,954 remote MCP endpoints, 38 more than on 21 October. None has opted out. Six have been claimed by their maintainers, three of them since 21 October.
Between 21 October, 20:00 UTC, and the end of October, the probe completed 122,452 runs against these endpoints, about 12,000 a day. Homepage trials are not in that number: a trial run is never stored.
What the latest run found
For each endpoint we take its most recent run in the seven days before the cut-off. 5,950 endpoints had one; the other 4 had been added but not yet probed.
| latest run | endpoints | share |
|---|---|---|
| signed, full: the server answered without credentials and every check ran | 3,527 | 59.3% |
| signed, auth-gated: the server asked for sign-in, and we signed what is observable without credentials | 1,936 | 32.5% |
| not signed | 487 | 8.2% |
Of the 487 unsigned runs, 456 did not reach a protocol revision: the server did not complete the MCP handshake, for example because it did not answer, or answered with something other than an MCP response. The other 31 completed the handshake but gave us no tool list we could fingerprint.
The shares moved little since 21 October. The auth-gated signing that started on 20 October now covers 1,936 endpoints, up from 1,900. What a signed attestation proves describes what the signature covers in each case.
Protocol revisions
3,560 of the latest runs completed the handshake, and the server told us which protocol revision it speaks. The other 2,390 did not: 1,934 auth-gated servers asked for sign-in before the handshake, and the 456 above never reached it. Two auth-gated servers completed the handshake and asked for sign-in afterwards; they are counted below.
| revision the server declared | servers | share |
|---|---|---|
| 2025-11-25 | 1,760 | 49.4% |
| 2026-07-28 | 1,060 | 29.8% |
| 2025-06-18 | 583 | 16.4% |
| 2024-11-05 | 84 | 2.4% |
| 2025-03-26 | 72 | 2.0% |
| other | 1 | under 0.1% |
2026-07-28, the newest revision, rose from 28.9% to 29.8% of the servers that told us their revision. The directory can be browsed by revision, for example servers declaring 2026-07-28.
Tool lists
3,789 servers had given us a tool list at least once by the end of October; those are the servers with a tool list in the directory. The median server among them exposes 7 tools in its latest list, as in the previous issue.
The hygiene checks changed
This is the change of the period. Between 27 and 31 October we revised our tool-description hygiene checks in several steps, from suite 0.17.0 to 0.19.2. The largest change is in 0.19.0: a long description is no longer flagged for its length alone. Tool descriptions that give orders explains what the checks look for.
Because the probe moved to suite 0.19.2 only on the evening of 31 October, the last day of the period, we read this number after the cut-off. On 4 November, 22:19 UTC, 26 of the 3,493 servers whose latest run evaluated the hygiene checks, 0.7%, had at least one tool description flagged. The previous issue reported 227 servers. The two numbers come from different rules and a different denominator, so they should not be read as a trend.
Tool-list changes
Between 21 October, 20:00 UTC, and the end of October, we recorded 902 tool-list changes on 492 servers. A change is a new tool list whose fingerprint differs from the server's previous one.
| changes in the period | servers | changes |
|---|---|---|
| 1 | 312 | 312 |
| 2 or 3 | 132 | 300 |
| 4 to 10 | 44 | 236 |
| more than 10 | 4 | 54 |
In 796 of the changes the new list had more tools than the previous one; in 90 it had fewer; in 16 the count stayed the same and the tools differed. From 22 to 31 October we recorded between 74 and 110 changes a day, about the level the previous issue ended on, while the list grew by only 38 endpoints. How often MCP toolsets change looks at changes in more detail.
How we counted
All numbers come from read-only queries against our production database, run between 21:50 and 22:20 UTC on 4 November 2026. The period is 21 October 2026, 20:00 UTC, to 1 November 2026, 00:00 UTC.
- The list: remote endpoints in our target table, by the time each was added. Claimed and opted-out counts are taken at the cut-off; the claimed count includes our own server, which we claimed as a paying user like any other maintainer.
- Runs: probe runs started in the period. Trial runs are not stored, so they cannot be counted.
- Latest run: for each endpoint, the most recent run that started in the seven days before the cut-off. Signed means the run has a signed attestation that is not disqualified; full and auth-gated are the coverage recorded in the signed payload. The unsigned reasons are the reasons stored with the run.
- Protocol revision: the revision the server returned in its initialize response in that latest run.
- Tool lists and median: each server's most recent stored tool list observed before the cut-off.
- Hygiene flags: each server's latest run on 4 November 2026, 22:19 UTC, under suite 0.19.2. A server counts in the denominator when that run evaluated the hygiene checks; servers that asked for sign-in or gave no tool list do not.
- Changes: consecutive stored tool lists of the same server whose fingerprints differ. We have not separated changes caused by a server returning a different first page of a paged list, so some changes may come from paging rather than from a server adding or removing tools.
Every observation here was made without credentials. A server may show a signed-in user different tools, or behave differently from another region. The methodology page describes the probe and the checks.