# ActualHumanTake

> Honest human product reviews, delivered in a target of 12 hours. A real person uses the product and writes the review. This is not an AI review service.

## What this service does

An agent or person submits a product and specific feedback focus. A human
reviewer uses the product and writes the feedback manually. Requests start
queued; a 12-hour turnaround is a target, not a guarantee. The service is free
during early access. It is not an automated review or testimonial generator.

## Agent discovery and connection

AI Catalog: https://actualhumantake.vercel.app/.well-known/ai-catalog.json
MCP Server Card: https://actualhumantake.vercel.app/api/mcp/server-card
MCP Streamable HTTP endpoint: https://actualhumantake.vercel.app/api/mcp
MCP setup guide: https://actualhumantake.vercel.app/docs

The MCP server exposes submit_product_for_review, get_review_status, and
get_public_stats. Connect to the remote HTTP endpoint, then list tools through
MCP. No API key is required during early access.

## API

Base URL: https://actualhumantake.vercel.app/api/v1

OpenAPI: https://actualhumantake.vercel.app/api/v1/openapi.json
Docs: https://actualhumantake.vercel.app/docs
MCP (Streamable HTTP): https://actualhumantake.vercel.app/api/mcp

No API key is required during early access. Rate limits apply.

## MCP

The remote MCP server uses Streamable HTTP at
https://actualhumantake.vercel.app/api/mcp. Tools:
- submit_product_for_review: queue a product for human review; private by default
- get_review_status: check status or retrieve feedback using id and receiptToken
- get_public_stats: read public service stats

The receiptToken is a secret. Human review remains manual.

### Submit a review

POST /reviews
Content-Type: application/json

{
  "product": {
    "name": "Acme Notes",
    "url": "https://acme.example",
    "category": "productivity",
    "description": "A minimal note app for developers with markdown and local-first sync.",
    "audience": "solo developers",
    "stage": "beta"
  },
  "contact": {},
  "submitter": { "type": "agent", "name": "Claude", "model": "claude-sonnet" },
  "focus": ["ux", "positioning", "pricing"],
  "publicOk": true
}

Response 201 includes:
- id
- receiptToken (returned once; keep it secret)
- statusUrl
- reviewPath
- slaHours
- feedback (null until completed)

Optional header: Idempotency-Key: <8-200 URL-safe chars>

The contact object is required. Email is required for humans and optional for
agents. Agents without email must keep the receipt token and poll for status.

### Poll status

GET /reviews/{id}
Authorization: Bearer <receiptToken>

Status values: queued, in_review, completed, rejected.
When completed, feedback.markdown contains the human-written review.
publicUrl is non-null only when the review is public and completed.

### Stats

GET /stats -> { completed, avgTurnaroundHours, slaHours }

## Limits

- 3 submissions per IP per hour, 10 per day
- 3 submissions per provided email per day
- 60 status reads per minute per IP
- Duplicate provided email + product URL within 24 hours returns 409

## Notes

- Private is the default. Private reviews are only readable with the receipt token.
- Public reviews get a shareable page at /r/{slug}.
- Never send passwords, API keys, or other secrets in the description. Use a temporary demo account if login is required.