# Finery

> Finery forges a business idea into a tested offer. Describe the idea in your own words; Finery interviews you, asking only for what it could not work out, then writes the Results: names, positioning, a pitch, CEO, CFO and tech question packs, and launch Checks.

Vocabulary: an Idea is one business idea being worked on; the Workshop is the interview that fills it in; a Forged Idea is one whose Results are written; Checks are the launch checks in the Results (domains, EU, UK and US trade marks, who already exists, relative search interest, UK companies), each with its source and date or why it was not checked; they are a first look, not legal clearance.

Price: EUR 9 per Forged Idea. Payment is not live yet: there is no checkout, and nothing is charged during the early preview.

## Pages

- [Homepage](https://finery.dev/): what Finery does, and the MCP URL to copy.
- [App](https://finery.dev/app): describe the Idea, answer Finery's questions (or let it assume the rest), forge it, and read the Results on the same page. Pages expose WebMCP tools to browser agents that support `document.modelContext`.

## Use Finery from an AI assistant (MCP)

- [MCP server](https://finery.dev/mcp): Streamable HTTP, no sign-in, no API key. Tools: `idea_start`, `idea_answer`, `idea_assume_rest`, `idea_forge`, `idea_get`, `idea_refine`, `idea_accept`.
- Claude: Open Claude and go to Customize, then Connectors. Choose Add custom connector. Paste https://finery.dev/mcp as the URL and add it. No sign-in is asked for. Ask Claude to forge an idea with Finery.
- Claude Code: run `claude mcp add --transport http finery https://finery.dev/mcp`

## MCP tools flow

Start, then answer or assume the rest, then forge, then get. The Idea id from `idea_start` is the only handle; keep it.

1. `idea_start`: Start an Idea from everything known about it, as free text (`context`). Returns what Finery understood, what is missing and up to three questions for the person.
2. `idea_answer`: Pass on the person's answers as free text (`idea_id`, `message`). Uses one round. Never invent answers for the person.
3. `idea_assume_rest`: Fill every missing answer with a labelled assumption (`idea_id`), when the person would rather not answer more.
4. `idea_forge`: Start writing the Results once the state is `ready_to_forge` (`idea_id`). Returns at once with state `forging`.
5. `idea_get`: Read the Idea (`idea_id`): state, open questions, rounds left, assumptions, and the Results once the state is `forged`. Poll it every few seconds while `forging`.
6. `idea_refine`: Write one part of the Results again (`idea_id`, `block`: names, positioning, pitch, ceo-pack, cfo-pack or tech-pack). Uses one round.
7. `idea_accept`: Accept the Results and close the Idea (`idea_id`). Cannot be undone; the Results stay readable with idea_get.

An Idea includes 6 rounds across answering and refining. Every answer carries `next`, which says what to do now.

## API

- [OpenAPI 3.1 spec](https://finery.dev/openapi.json): every REST route under `/api`, with request and response schemas, auth schemes and error bodies.
- [Full agent guide](https://finery.dev/llms-full.txt): this file plus the REST flow and auth, inline.

## Optional

- [Sitemap](https://finery.dev/sitemap.xml)
- [Canvas screen](https://finery.dev/app/canvas): the earlier field-by-field editor for the same Idea.