cairn

cairn

Repo-native task management. One task graph your agents and you share.

·

·

·

·

cairn is a task graph that lives as Markdown files in your repository. A single Go binary

(`cairn`) serves it to AI agents over MCP and to humans over

a clean web UI. One source of truth: the files. One rule-set: `internal/task`. No database.

• Lives in your repo. A task is Markdown (YAML frontmatter for the machine, a prose body for

humans). Tasks branch, merge, and review like code; a task's history *is* its git history.

• Built for agents. Agents list ready work, claim it, run an observable session with

heartbeats, and hand off for review. Every write is stamped with the actor that made it.

• One rule-set, two front-ends. The web UI and the MCP server are thin adapters over the

same engine, so an agent and a human always see the same gates and the same readiness.

• Connect any agent in one click. The Connect page detects installed agents and writes their

MCP config for you, each under its own identity.

Quickstart

make build              # -> bin/cairn
cairn web --repo .      # open the board in your browser

Then open the Connect page and wire up an agent in one click, or do it by hand:

claude mcp add cairn -- "$(pwd)/bin/cairn" serve --actor agent:claude --repo "$(pwd)"

Full walkthrough: **Installation** →

**Quickstart**.

Run it your way

Supported agents

One-click connect for Claude Code, Cursor, Codex, Windsurf, OpenCode, Kilo Code, and Pi;

a copy-paste guide for Antigravity and any other MCP client. See the

Agents guide.

Development

make check       # gofmt + go vet + go test ./...
make web-dev     # Vite dev server (proxies /api -> :8080)
make desktop-dev # native window against a dev server

This repo dogfoods cairn: its own work is tracked in `.cairn/`. See

AGENTS.md and .cairn/WORKFLOW.md.

Security

cairn is a local, single-user tool with no authentication by design. See

SECURITY.md for the trust model and how to report a vulnerability.

License

MIT.