Kanon

Project management, rebuilt for the age of AI coding agents.

Why Kanon?

Jira, Linear and Trello were designed for humans clicking in browsers. But more and more of the code that ships today is written by AI agents — and those agents cannot drive a board that was never built for them.

Kanon treats AI agents as first-class users, not as a bolted-on integration. Issues, sprints, transitions, dependencies — all operable natively through the Model Context Protocol (MCP). The web UI is there for humans. The MCP server is there for agents. Both talk to the same source of truth.

The name

Kanon (Greek: κανών, kanṓn) — "the rule, the measuring rod". In ancient Greek it referred to the straight standard against which every other measurement is judged. Fitting, for a system whose job is to set the cadence of engineering work.

What you get

• MCP-native — your AI assistant (Claude Code, Cursor, Antigravity) creates, transitions and comments on issues directly, with no custom glue code.
• Web Kanban — a fast React 19 UI for the moments you want to drive the board yourself.
• Self-hosted — your data stays in your Postgres, your infra, your network.
• SDD-aware lifecycle — issues move through states that mirror real work: backlog → explore → propose → spec → design → tasks → apply → verify → archive.
• Zero-friction AI setup — one command (npx @kanon-pm/setup) configures your AI tools with MCP, skills and workflows.
• Real-time — WebSocket-backed updates across clients.

Inspiration

I built Kanon because my own workflow broke.

I was writing specs, having agents implement them, verifying output, archiving changes — and the tools I was using to track that work (Notion boards, Linear, custom Markdown trees) could not talk to the agents doing the work. Every hand-off needed me to translate between the board and the agent.

Kanon started as a single-user tool to close that gap — a board where the agent can create its own issue when it spots a TODO, update state when it finishes a phase, and link dependencies without being asked. It is still opinionated toward that workflow, and it is the laboratory where I keep refining how humans and agents share a project.

Quick start

Requires Node.js 20+, pnpm and Docker.

git clone https://github.com/mmmarxdr/kanon.git
cd kanon
pnpm bootstrap   # install deps, generate prisma, migrate, build mcp + setup
pnpm dev:start   # boot Postgres, API, web and Engram

Open http://localhost:5173 and sign in with dev@kanon.io / Password1! (workspace: kanon-dev).

To wire up your AI tools:

npx @kanon-pm/setup

For advanced install options (manual Postgres, no-Docker setup, CI scripting), see docs/INSTALL.md. For AI tool setup details and supported clients, see docs/AI_TOOLS.md.

Tech stack

Fastify 5, Prisma, PostgreSQL, React 19, Vite 6, TanStack Router + Query, Tailwind CSS 4, Zustand, @modelcontextprotocol/sdk, Vitest, Playwright. Monorepo with pnpm workspaces.

Project layout

For dev commands, build/test/release instructions and project internals, see docs/DEVELOPMENT.md.

Status

Early development — v0.2.0. Expect breaking changes until v1.0. Heading toward v1 with hardened multi-user support, a public API key flow and a stabilized MCP surface.

Contributing

Contributions are welcome. Read CONTRIBUTING.md before opening a PR.

License

Kanon is licensed under the Apache License 2.0.

Copyright © 2026 Marc Dechand. See NOTICE for attribution requirements.

You are free to use, modify, and distribute Kanon — including for commercial purposes — provided you preserve the copyright, license, and NOTICE files in any redistribution. Apache 2.0 includes an explicit patent grant that protects both users and contributors.