agent-gate MCP server

agent-gate is a helper that makes your AI prove its work before it tells you it is finished. Instead of just saying "done", the AI has to pass a checklist of checks, and every decision gets written into a log that cannot be quietly changed. It is handy for anyone who uses AI agents to do real tasks and wants a bit more proof that the work is actually complete.

What this MCP server does

You ask your AI to finish a task, and before it says "done" it calls agent-gate. The helper hands back a short checklist of things that must be true, like "tests passed" or "no secrets leaked". Your AI fills in what it knows, and agent-gate checks each item strictly: if something is missing, it counts as failed, not passed. When the AI makes a decision, agent-gate writes it into a running log with a special fingerprint, so if anyone edits or deletes an old entry, the log can tell. At the end you can read the log and see whether the chain is still intact.

Click to zoom

What you can do with it

• Ask your AI to check its work against a ship checklist before saying it is done
• See exactly which checks are still blocking the work
• Record each decision your AI makes as an honest receipt
• Read back the full receipt log and confirm nothing was changed
• Keep a tamper-evident trail of what your AI decided and why
• Require human approval for irreversible or outward-facing actions

Try asking your AI

• “Before you tell me this is finished, run the ship checklist and show me what is still blocking.”
• “Record a receipt for shipping version 0.2 with tests passing.”
• “Read back all the receipts and tell me if the chain is still intact.”
• “Check whether my last change passes the gate, and if not, list the missing items.”

What it gives back to you

You get back a short answer: whether the gate passed, and a list of any checks that are still blocking. When a receipt is recorded, you see the entry with its sequence number, decision, verdict, and a long fingerprint hash. When you read the log, you get every receipt plus a simple true or false telling you whether the chain is intact.

Before you start

Who it's for

People who run AI agents on real tasks and want a simple, honest way to check the work before it is called done.