56% of the 293 go to market tools counted here have an MCP server of any kind. Applying the second filter, that you can actually get an account without a contract, 117 entries pass both tests, which is 40% of the directory.
| Filter | Entries left | Share of 293 |
|---|---|---|
| Every tool counted | 293 | 100% |
| Has a server of any kind | 165 | 56% |
| Server is first party | 144 | 49% |
| Server, and free or paid self serve | 117 | 40% |
| Server, first party, and free to start | 42 | 14% |
| And somebody here has actually run it | 0 | 0% |
Counted 2026-08-25 from directory.json. The last row is the honest floor of this whole exercise and it is published on the front page rather than buried here.
Every percentage above the last two is a documentation reading. It says a vendor published something and a check found it on a date. The bottom row says somebody ran the thing. This directory's bench tested count is 0, it stays there until Andrew personally runs a tool on a stated date, and no vendor can buy a change to it.
Do not read 56% as a permanent state. It is a reading taken on 2026-08-25 from 293 entries whose facts were pulled by hand on 2026-08-24, 2026-08-25. The point of stamping every number with a date is that the next reading is comparable to this one.
Every number on this page is generated from directory.json at build time and carries the date it was baked: 2026-08-25. Nothing is typed by hand, nothing is rounded, and nothing is estimated. The underlying data is published in full. Where the honest answer is a zero, the zero is printed.