SayDo MCP server

╔════════════════════════════════════════════════════════════════════════════════════════════╗
║                                                                                            ║
║                         ███████╗ █████╗ ██╗   ██╗██████╗  ██████╗                          ║
║                         ██╔════╝██╔══██╗╚██╗ ██╔╝██╔══██╗██╔═══██╗                         ║
║                         ███████╗███████║ ╚████╔╝ ██║  ██║██║   ██║                         ║
║                         ╚════██║██╔══██║  ╚██╔╝  ██║  ██║██║   ██║                         ║
║                         ███████║██║  ██║   ██║   ██████╔╝╚██████╔╝                         ║
║                         ╚══════╝╚═╝  ╚═╝   ╚═╝   ╚═════╝  ╚═════╝                          ║
║                                                                                            ║
║                               does the tool do what it says                                ║
║                                                                                            ║
╚════════════════════════════════════════════════════════════════════════════════════════════╝

DRAFT COPY — pending the owner's pass. Everything below is measured or mechanical, and none of it is a marketing claim, but the wording has not been approved.

A publisher declares, in a signed machine-readable contract, exactly what a tool does. A harness makes the tool run and checks it against that contract. Every run emits a hash-chained receipt anyone verifies in a browser, with no account, and without trusting the issuer.

The question it answers is narrow on purpose: does this tool do what it says? Not is it safe, not is it good. A tool that refuses to act has not been shown to be safe — it has not been shown anything, and SayDo says that in those words rather than returning a green tick.

What a verdict can be

pass the declared behaviour held under the run fail the tool did something it declared it would not not-covered the run established nothing either way

not-covered is never quietly upgraded to pass. Most of the interesting failures in this domain come from silence being read as good behaviour.

And a whole run resolves to one of four things, not two:

CONFORMANT it did what it said, and it was shown doing it NOT CONFORMANT it did something it declared it would not INCONCLUSIVE nothing failed, and nothing was established (revoked / expired, once a registry has spoken)

INCONCLUSIVE carries most of the weight. A server that declines every call — the commonest kind in the public sweep — makes no network request, writes no file and starts no process, so a naive harness marks all three as passes and issues a warrant for a program that did nothing. SayDo requires a run to have demonstrated conduct before any of those checks can pass, and a server describing itself does not count as conduct.

In your CI

with: command: python -m my_server declaration: saydo.declaration.json

• uses: vince-gonzalez/saydo@main

The build fails if the tool breaks its declaration, and also if the run established nothing — because a green check that proves nothing is worse than no check, being a claim of conformance that nobody actually made. Set require-coverage: false to let such a build through; the receipt still records it as inconclusive and never as a warrant.

This is the place the tool is worth the most. Coverage comes from the tool actually doing its work, which needs the credentials and inputs it accepts — and those exist in the publisher's own CI and essentially nowhere else.

Run it on anything

saydo puts one tool under test and writes a receipt. The tool does not have to be yours, and it does not have to be an F-Keys one:

Shortened. The full README is on GitHub.