Reelier MCP server

Reelier is a tool that watches an AI agent do a job once, saves that exact run as a reusable recipe, and then replays it later without asking the AI to think again. It also writes a short receipt showing what ran and what changed. It is handy for developers and teams who want proof that an agent still does what it used to do.

What this MCP server does

You ask your AI to record a job that an agent just did, for example a set of tool calls that worked. The AI passes that request to Reelier, which saves the run as a skill file. Later you ask the AI to replay that skill, and Reelier runs it again step by step without calling a model, so it costs no tokens. You can also ask it to compare two runs, and it tells you whether each step is the same or has drifted. Every replay comes back as a receipt showing what ran and what changed.

Click to zoom

What you can do with it

• Record a working agent run and freeze it as a replayable skill
• Replay a saved skill without calling an AI model, so no tokens are used
• Compare two runs and flag which step drifted
• Inspect what an agent can see and what it might change before you let it write
• Convert an existing Agent Skill into a Reelier skill
• Write a CI workflow that checks for drift and adds receipts to pull requests
• Push receipts to a ledger so you get a permalink and a verified badge

What it gives back to you

You get back a short receipt in the chat: what ran, which files or fields changed, and whether anything was flagged. For a diff, you get a per-step answer of SAME or DRIFTED, with the failing check as the reason. Skill files and traces are saved on your machine under a .reelier folder, and you can open them like any other file. Receipts can also be pushed to a ledger, which gives you a link you can share.

Before you start

Who it's for

Developers, open source maintainers, and teams who run AI agents and want a repeatable test and a proof of what changed.