Human Approval - HITL Platform

@human-approval create a loop called 'Production Deploys' and invite sarah@example.com and mike@example.com as reviewers

Loop created successfully with ID loop_abc123. Sarah and Mike have been sent invitation links to join as active reviewers. The loop is now ready to receive approval requests. Your account remains as the loop owner and can broadcast requests once at least one invitee accepts their membership.

This example shows loop creation, the foundation for all approval workflows. The MCP returns invitation links but cannot force acceptance — you must wait for humans to join before the loop becomes functional. Requires API_KEY auth with loop creation permissions.

@human-approval send an approval request to the 'Production Deploys' loop asking if we should merge PR #847 to main, with a 2-hour timeout

Approval request req_xyz789 broadcast to 2 active members in 'Production Deploys' loop. The request asks: 'Should we merge PR #847 to main?' Reviewers have 2 hours to respond via their notification channels. This consumed 1 request from your provider allowance (47 remaining). Poll the request status with get-approval-request to see human responses as they arrive.

This demonstrates the core human-in-the-loop capability: pausing AI workflows for human judgment. The MCP does not block or wait — your agent must poll for responses. Each request consumes allowance quota. If no human responds before timeout, the request expires without approval.

@human-approval check all pending approval requests in the last 24 hours and summarize which ones are still waiting for human review

Found 3 approval requests from the past 24 hours: req_xyz789 (PR #847 merge) is pending with 0/2 responses after 45 minutes; req_abc456 (database migration) was approved by 2/2 reviewers 6 hours ago; req_def012 (API key rotation) expired without response after 2-hour timeout. One request still needs human attention before your workflow can proceed.

This synthesis example shows how an AI can monitor approval state across multiple requests and surface blockers. The MCP returns historical data but cannot predict when humans will respond. Useful for daily standup summaries or identifying stalled workflows that need follow-up.

When your AI assistant needs a human gate for spend decisions

A 6-person startup uses an AI agent to draft vendor contracts and schedule payments. The agent can negotiate terms and queue invoices, but any spend over $500 needs a human yes-or-no before the wire goes out. This MCP is built for exactly that gate: the agent calls Create Approval Request mid-workflow, waits for a loop member to approve or reject, then continues or halts based on the response. The 10-tool surface is narrow by design—this isn't a collaboration platform, it's a blocking checkpoint. If your agent workflows already have clear decision thresholds and you need audit trails for compliance, this is the right primitive. If you're looking for threaded discussion or multi-stage review, you'll outgrow it fast.

Gating AI responses when the stakes get legal or reputational

A 3-person SaaS support team uses an AI agent to draft replies to inbound tickets. Most answers ship automatically, but refund requests over $200 or any mention of 'lawyer' trigger a human approval loop before the reply goes out. The agent creates the request, a support lead reviews the draft in Slack or email (via the invitation link), and approves or rewrites. The MCP's Get Approval Request tool lets the agent poll for the decision without blocking other tickets. The trade-off: if you need reviewers to leave inline comments or suggest edits, this MCP won't carry that payload—it's a binary gate, not a collaboration surface. For high-volume support teams where most escalations resolve in under 5 minutes, the request allowance and loop-member management become the bottleneck to watch.

When your content agent needs a final human read before publish

A 4-person marketing team runs an AI agent that drafts blog posts, social captions, and email newsletters on a weekly cadence. The agent writes, formats, and stages everything in the CMS, but a human editor must approve each piece before it goes live. The agent calls Create Approval Request with the draft link, the editor reviews in their own tool, and the agent polls Get Approval Request to decide whether to hit publish or archive. This works cleanly when the review is pass-fail and the editor's feedback happens outside the loop (in Google Docs comments, for example). If your editorial process requires the reviewer to leave structured notes or request revisions inside the approval flow, this MCP won't carry that context—you'd need a richer integration or a custom webhook setup.