Can my app or AI tool ask real people?

Yes, in a sandbox today. Developers can ask through a REST API and an MCP server, and each answer comes with a signed receipt. The sandbox uses test members and nothing is charged. Keys are issued by invitation during early access. See the Developers page for details.

Can I use it with ChatGPT?

Not through ChatGPT's connector screen yet, because it offers OAuth or no sign-in and can't send our API key. A Custom GPT Action over our REST API works, as do the OpenAI Responses API and Agents SDK.

Is there a free tier?

The sandbox is free and always will be. Live pricing is per answer, by answer type: a quick take is from 30p a person plus 50p a question (launch example), and early access is free.

Keys are issued by invitation during early access. Use the developer route on the contact page with your name, an email address, what you're building and roughly how many questions a week you expect.

What do sandbox answers look like?

Canned lines from synthetic test members, marked synthetic: true, usually within a few minutes. Use them to test your integration, not to learn anything about people.

Not yet. The API is plain HTTP with an OpenAPI 3.1 document you can generate a client from, and the MCP server works with any MCP client. Runnable Node examples cover REST, MCP and receipt verification.

What uptime can I expect?

We don't publish an uptime commitment during early access. Design for retries: send an Idempotency-Key when asking, and resume event streams with Last-Event-ID.

Can my agent pay per call in crypto?

Not yet. Keyless pay-per-call (x402) is refused today; every call needs an operator key.

Who do I contact about a problem?

Use the developer route on the contact page and include the requestId from the error response. A person reads every message.

Developers: REST and MCP for human answers

Test a human-feedback workflow from your code or AI tools. The invitation-only sandbox returns synthetic answers and signed sandbox receipts today; live human API answers are planned.

Overview

The planned human service returns a choice or rating, a line of why and a summary once the pool closes. Today, test that question-and-answer workflow with synthetic sandbox members. Each sandbox answer carries a signed receipt you can verify against the sandbox public key; it is not evidence that a person answered.

Start with the quickstart. It takes a sandbox key and about five minutes.

Start here

QuickstartYour first sandbox question in about five minutes, over REST or MCP.
Connect an AI assistantClaude, ChatGPT and OpenAI tools, Codex, Cursor, VS Code and others.
API referenceEvery REST operation, from the OpenAPI 3.1 document.
ReceiptsCheck offline that a person answered.

Giving an autonomous agent access? Read the rules and controls for AI agents first.

What you can ask

Pick one of two or three options, with one line of why.
Rate something from 1 to 5, with one line of why.
A first impression of an image or a short piece of text, in one line.

Questions are images (PNG, JPEG or WebP) or text up to 280 characters. Links are refused for now. Pools run from 5 to 50 people.

How it connects

REST API at https://pennyforyours.com/v1, with an OpenAPI 3.1 reference.
MCP server at https://pennyforyours.com/mcp, so AI assistants and agents can ask as a tool call. Set it up in Claude, ChatGPT and OpenAI tools, Cursor, VS Code or Codex.
Results arrive by long-polling (wait up to 25 seconds for news) or a resumable event stream.
Every call uses an operator API key with an accountable owner. There is no keyless access.

Machine-readable entry points
What	URL
MCP endpoint	`https://pennyforyours.com/mcp`
REST base	`https://pennyforyours.com/v1`
OpenAPI 3.1	`https://pennyforyours.com/v1/openapi.json`
Receipt keys (JWKS)	`https://pennyforyours.com/.well-known/jwks.json`
Sandbox receipt keys	`https://pennyforyours.com/v1/sandbox/jwks.json`
Summary for AI tools	`https://pennyforyours.com/llms.txt`

Every answer carries a signed receipt (JWS, ES256) that binds the exact question, the exact image or text shown, the answer and its status. How receipts work. Pricing is the same per-answer pricing as the website: a quick take is from 30p a person, plus 50p a question for screening and the summary (launch example). Early access is free, and the sandbox is always free. Pricing by answer type.

Status

Sandbox only. The REST API and MCP server run in sandbox mode: keys start pfy_test_, synthetic test members answer within a few minutes, and nothing reaches real people or costs anything. Keys are issued by invitation during early access. Sandbox and limits.

What is live and what is not
Area	Status
Sandbox keys (pfy_test_)	Live, by invitation
Live keys (pfy_live_) and charging	Not yet
Question types over the API	Quick takes only: pick one, rate 1 to 5, first impression
Results	Long-poll (up to 25 seconds) and Server-Sent Events (resumable)
Webhooks	Not yet
OAuth for MCP clients	Not yet; API keys only
Keyless pay-per-call (x402)	Not yet; every call needs an operator key
Receipt anchoring on a public ledger	Not yet; receipts are signed, and anchor is null

Developer promises

PennyForYours is built and run in the UK by David Bryan. The API and MCP server are part of the same service as the website, with the database and server code in London. These docs are generated from the same source as the API wherever possible: the API reference from the OpenAPI document and the MCP tool list from the server's own tool definitions, so they cannot drift apart.

Compatibility. The REST API is versioned in the path (/v1), and receipts carry a schema version (v). Within /v1 we may add fields, operations and MCP tools, but we will not remove or rename them.
Notice of breaking changes. A change that would break existing calls goes into a new version. We email key holders, at the address they gave when they asked for a key, before such a change affects anything they use.
Business use only. API and MCP access is for organisations and developers acting for their trade, business or profession, under an operator key with an accountable owner. The terms say the same.
A changelog. Every change to the API, the MCP tools or the receipts is listed below with its date.

Rules for API use

Members are told when a question came through an AI tool.
Not allowed: regulated advice (medical, legal, financial), collecting personal data, asking for or writing reviews, targeting a real person, CAPTCHA solving, scam testing, or anything unlawful.
Neutral questions. Don't lead people towards an answer. Every live question is screened.
Answers are for your decisions, not for publishing as reviews or testimonials.

Data and security

Keys are stored as a hash; the full key is shown once.
Rate limits are per key and shared between REST and MCP.
Errors are RFC 9457 problem documents.
Images you upload have hidden metadata stripped before anyone sees them; the receipt's ah is the hash of what was shown.
Our database and server code are in London; uploaded images are stored by Netlify in the US. We do not train AI models on your questions or the answers. Privacy in plain English.
Built on maintained open-source components under permissive licences, including the official MCP TypeScript SDK and jose for receipts. The tech underneath.

Changelog

Changes to the API, the MCP tools and receipts
Date	Change
5 October 2026	Added: optional `offer_pence`, `required_tier`, `topics` and `expected_minutes` when quoting and asking; `GET /v1/topics` and the `list_topics` MCP tool. The receipt check `time_recorded` is now confirmed against our server's clock. Nothing removed or renamed.
4 October 2026	Sandbox opens: REST `/v1` with an OpenAPI 3.1 document, the MCP server at `/mcp`, and signed receipts. Sandbox receipts are signed with a separate key and never verify as proof of a person. Image uploads up to 4 MB.

Talk to us

Request a sandbox key or tell us what you're building through the developer route. Include an email address: we use it for notice of changes.
Report a bug or a docs error the same way, with the requestId from the error response.
Security issue? Report it privately through the help and support form with “Security” at the start of the subject.

Ask real people from your code and AI tools