commitlayer
CommitLayer checks what AI agents actually did. An agent reports that it refunded an order, sent a payment or cancelled a subscription; CommitLayer reads the systems the action touched afterwards (Stripe, Shopify, your CRM, your own events) and returns PASS, FAIL or UNKNOWN with the step that did not land. The agent’s own report is never the evidence. With this connector, Claude can: List recorded actions and their verdicts, newest first, filtered by contract, workflow, verdict or month Open one action’s full verification: both lenses, status, reason codes, the checks that ran and pointers to the evidence See what is still pending, which system of record it is waiting on and when it settles Summarise a month’s verdicts per contract Read the workspace’s contracts, validate a contract change and dry-run it against a month without saving anything Draft an action contract from what was actually recorded Check connector health and the workspace’s plan and usage Verify a statement, verdict or dispute hash against the ledger Record an action before it is performed, when the token carries the actions:record scope Everything else is off limits by design. The connector never writes a fact, changes a verdict, activates a contract, touches a connected system or deletes anything. Fourteen of the fifteen tools are read-only; the one write records an intent and is not destructive. Sign in with your CommitLayer account and approve the connector for one workspace with the scopes you choose. Tokens expire after an hour and refresh for thirty days; leaving the workspace ends them. You need a CommitLayer workspace, and the free tier is enough.
Details
https://api.commitlayer.ai/mcp · streamable-http
