Getting Started

What is IntentDocs?

Spec-driven development for AI coding agents — story maps, acceptance criteria, and a CLAUDE.md context harness kept in sync with your codebase.

Spec-driven development for AI coding agents

IntentDocs is an MCP-native spec-driven development platform for Claude Code, Cursor, and other AI coding agents. It captures product intent as structured specifications — a story map with acceptance criteria, a data model, and a recommended build order — and gives your coding agent live access to those specs, either over MCP or through the intent CLI.

Instead of a founder or engineer re-explaining context in every chat, the agent reads it directly: what to build, why, in what order, and how to prove each piece is actually done.

The context engineering harness

IntentDocs isn't just a story tracker. It's the structured context an AI coding agent needs to work reliably on a real codebase, organized into six kinds of context:

Context type	What it holds	How to populate it
Instructions	Role, goals, and boundaries for the project	Project page → Instructions
Scope	Story maps, user stories, acceptance criteria	Web UI, or `intent add` / `intent propose`
Memory	Architectural decisions, lessons learned	`intent decide` / `intent done`
Examples	Reference patterns — before/after code	`intent example`
Tools	APIs, services, MCP configuration	Project page → Tools
Guardrails	Hard constraints, safety rules	Project page → Guardrails

All six are assembled into a single CLAUDE.md (or AGENTS.md / GEMINI.md) file with one command:

intent sync --context

intentdocs.com                        Your project
┌─────────────────────┐               ┌──────────────────────────┐
│  Project page        │               │  CLI                     │
│  ├ Instructions       │               │  ├ CLAUDE.md   ← harness │
│  ├ Story maps         │◄── sync ─────►│  ├ INTENT.md             │
│  ├ Decisions          │               │  ├ .intent/              │
│  ├ Examples           │  --context    │  │   stories/<id>.md     │
│  ├ Tools / MCP        │──────────────►│  └ mcp.json              │
│  └ Guardrails         │               │                          │
└─────────────────────┘               │  Claude Code / Cursor    │
                                       └──────────────────────────┘

What you get

A story map — personas, activities, and stories, each with a priority (must / should / could), an effort estimate, and acceptance criteria.
INTENT.md — a plain-markdown snapshot of the story map, generated by intent sync, that lives in your repo and travels with your code. See Story maps and INTENT.md.
Acceptance criteria your agent can prove — some criteria are executable shell commands your agent (or intent check) can run and get a real pass/fail from, instead of taking the agent's word for it.
An MCP server — intent mcp exposes your story map as tools and resources any MCP-compatible agent can call natively, no copy-pasting context into a prompt. See MCP Integration.
A CLI — intent — for everything else: picking up the next story, marking work done, logging decisions, and keeping the local harness in sync. See CLI Reference.

Where to go next

New to IntentDocs entirely? Continue to Installation.
Already have a project on intentdocs.com? Skip to Connecting a repo.
Wiring up Claude Code or Cursor? Go straight to MCP Integration.