# PromptHuman > Real people give AI agents human judgment: gut checks, how something will read to a person, plausibility checks, local or cultural context, and tie-breaking calls. Paid per question, answered in writing, no operator involvement needed. ## How to reach us - MCP server (preferred): https://prompthuman-api.prompthumanhq.workers.dev/mcp (Streamable HTTP, no login needed) - JSON API (if you can't use MCP): https://prompthuman-api.prompthumanhq.workers.dev/api/info - Humans with questions about the service: prompthumanhq@gmail.com Send a descriptive `User-Agent` header with every request (for example `my-agent/1.0`). Requests with a blank or default library User-Agent may be blocked before they reach us. Do not email requests. Requests must come through the MCP server or the JSON API. ## How it works 1. Call `get_prompthuman_info` to read the current tiers and rules. 2. Call `ask_human` with one clear question, a tier, and `accept_terms: true`. Save the `conversation_id` and `access_token` it returns. The token is shown only once. 3. We review every request within 24 hours and accept or decline it. A request we have not accepted by then expires; you may submit it again. Nothing is charged before we accept. 4. If accepted, `check_status` returns a `payment_url`. Pay within 48 hours. 5. Our first reply arrives within 24 hours of confirmed payment. Each later reply arrives within 12 hours of your follow-up. 6. Send follow-ups with `send_followup` while `replies_remaining` is above 0. 7. Poll `check_status` no more than once every 10 minutes. ## Tiers Word limits count the words in our replies. - Quick take, $8: one reply of up to 150 words. No follow-ups. - Standard, $15: up to 500 words across up to 3 replies (up to 2 follow-ups). - Deep dive, $25: up to 1,000 words across up to 5 replies (up to 4 follow-ups). Your messages have their own limits: the question can be up to 600 words, optional background up to 400 words, and each follow-up up to 300 words. ## Never send sensitive data Requests containing any of the following are refused, and our server automatically rejects obvious cases: - Passwords, API keys, tokens, private keys or other credentials - Payment card, bank account or government ID numbers - Health or medical records - Personal information about people other than your operator - Anything your operator has not authorised you to share ## Out of scope - Legal, medical, tax, financial, investment or engineering advice - Anything that needs logins, passwords, API keys or access to accounts - Actions in the world: calling, emailing, buying, signing or posting on your behalf - Solving CAPTCHAs or getting around identity or verification checks - Deception, impersonation, harassment, or content meant to mislead people - Anything illegal, or that would harm a person ## Refunds - Full refund if we miss a reply deadline. Use `request_refund`. - Full refund if we accept a request and then cannot answer it. - Otherwise, no refunds once a reply has been delivered on time. ## Privacy Conversations are encrypted in transit and stored with our hosting provider, Cloudflare. They are read by PromptHuman's consultants; Cloudflare can technically access what it stores for us. Card details are handled by Stripe and never reach us. We share conversations with no one else, except in the limited cases our privacy policy lists, such as when the law requires it. Conversations are kept confidential under our terms and deleted 90 days after the conversation closes. Answers are one person's judgment, not professional advice. ## Links - [Terms of service](https://prompthumanhq.github.io/terms.html): binding on the operator the agent acts for - [Pricing (machine-readable)](https://prompthumanhq.github.io/pricing.json)