MCP
AI agents · Model Context Protocol
Let Claude, Cursor or any MCP agent pause for a human review mid-task.
In plain English
LoopQuest is a place where a person quickly checks the work an AI or automation has done. Connecting MCP means that whenever MCP produces something that should be looked at by a human, it sends that item to LoopQuest. Someone on your team reviews it in a few seconds, and their decision is sent straight back to MCP so the rest of your process can carry on, with a real person accountable for the call.
When you'd use it
Your agent is about to do something consequential — send an email, issue a refund, merge a PR, delete a record. Instead of acting blind, it calls the LoopQuest tool, a human makes the call in a quick game, and the verdict returns to the agent before it proceeds.
What you'll need
- A free LoopQuest account. Sign up if you don't have one.
- A project and its API key — find it under Workspaces → your project → API keys. The key is how MCP proves it's allowed to send you work.
- An account on MCP itself, with the workflow or agent you want to add a human check to.
How to set it up
- 1Add the server to your MCP client config (Claude Desktop, Cursor, or your own runtime).
- 2Set LOOPQUEST_KEY to a project API key from Workspaces → your project → API keys.
- 3Restart the client — the agent gains two tools: create_review_task and get_task_status.
- 4Prompt the agent to request a human review before risky actions; it polls get_task_status for the verdict.
Copy-paste starting point:
// claude_desktop_config.json (or your MCP client's config)
{
"mcpServers": {
"loopquest": {
"command": "npx",
"args": ["-y", "loopquest-mcp"],
"env": { "LOOPQUEST_KEY": "lq_..." }
}
}
}Recipes: gate or monitor
Two ways to put a human in the loop with MCP. A gate blocks until someone approves; a monitor reviews in the background without slowing anything down.
Gate — block until approved
MCP can pause a run and resume it when the verdict arrives, which makes it a clean blocking gate.
- 1In your MCP flow, send the output to LoopQuest with the mode set to "gate", plus a callback so it can be resumed.
- 2MCP pauses at that step (a Wait node or resume URL) instead of carrying on.
- 3A reviewer approves or flags it; LoopQuest calls back and the run picks up exactly where it left off.
- 4Branch on the verdict, and set a timeout so a missing decision never blocks the flow forever.
Monitor — review without blocking
Monitor mode never holds MCP up. The flow proceeds immediately and LoopQuest reviews a copy for quality.
- 1Send the output with a mode of "monitor" and don't wait on the result.
- 2MCP carries straight on, nothing is blocked.
- 3Verdicts are logged and scored over time. Sample a fraction for QA, or review every item.
What happens after a review
The moment a reviewer decides, LoopQuest sends the result back to the web address you gave it (your callback_url). The message says what was decided, whether it was approved, flagged, which option was picked, or the corrected values, so your automation can act on it automatically.
For developers: each callback is signed with an X-LoopQuest-Signature HMAC header. Verify it before trusting the payload. Full details are in the API reference.
Pairs well with
The review games that best fit the kind of output MCP tends to produce.