What is IntentDocs?

IntentDocs is a spec-driven development tool for Claude Code, Cursor and other AI coding agents. You describe your product in plain words. IntentDocs turns that into a story map with acceptance criteria. Your agent reads the spec over MCP, builds it story by story, and reports which criteria pass. Nothing is approved until you say so.

How it works

The product follows four steps: Describe, Specify, Build, Prove.

Step	What you do	What you get
Describe	Answer five short questions about your product, or let your agent read an existing codebase.	A vision: who it's for, the problem, what makes people switch, the first release, and how you'll know it's working.
Specify	Review the story map IntentDocs drafts from your vision, and edit it.	Stories grouped by activity and release, each with acceptance criteria: what "done" means, written so you can check it.
Build	Connect your coding agent over MCP and put the stories in build order.	The agent claims a story, plans it, builds it, opens a pull request and links it to the story.
Prove	Read the verdict on each story, then approve it or send it back.	Every criterion reported as passing or failing, the tests that prove each one, and a record of who approved what.

The spec is also saved as INTENT.md, a plain markdown file in your repository. It is an open format and works without an account. See Story maps and INTENT.md.

The parts

• The web app at intentdocs.com. Each project has four tabs: Overview, Spec, Plan and Validate. This is where you write the spec, order the work and review what agents built. See Using the web app.
• The MCP server at https://www.intentdocs.com/api/mcp. Your coding agent connects to it to read stories, save plans and tests, and report results. See Connecting your agent.
• The CLI, npx intentdocs. It connects a repository to a project, keeps INTENT.md current, sets up your coding agent and turns on CI test reporting. See Installing the CLI.

Who does what

Your agent can draft stories, write plans, build, save tests and report results. It can mark a story dev complete once every must criterion passes. It cannot approve its own work: approved is set by a person in the web app.

You decide what gets built, and when. A project can require a person to mark a story ready, or to approve the agent's plan, before any agent builds it. See Working with agents.

Where to go next

1. Create a project and answer the vision questions.
2. Connect your agent over MCP.
3. Install the CLI and connect your repository.

Already have a codebase? The brownfield agent prompt has your agent read it and write a report that seeds your story map with what's already built.