tick-md, an open-source multi-agent task tool built by Purple Horizons
Open-source task coordination for AI agents and the people who work with them. One Markdown file, tracked by Git.
Services Provided
tick-md is our open-source tool for coordinating work between AI agents and the people who work with them. Tasks live in one file, TICK.md, inside your project. Agents and people claim, update and finish tasks through a command line tool or an MCP server. Every claim, status change and completion is a Git commit. We built it because we wanted our own AI agents to share one task list that people could read too.
It is built for development teams, AI agent swarms and solo developers. tick-md is MIT licensed and live on GitHub and npm, and the latest npm release shipped in February 2026. To try it, run npm install -g tick-md, then tick init in your project.
For the longer story, including a walk-through of our assistant Mia's daily workflow, read our blog post on tick-md. The code is on GitHub, the CLI is on npm, and the website and docs are at tick.md. If you want help putting agents to work on your team, see our AI agent systems service.
Challenge
- With more than one AI agent on a project, each one needs to know which tasks are taken, which are blocked and which are done.
- Agents often work in separate sessions and tools, so they need a shared task list they can check at any time.
- Typical task tools are built for people first, with logins, web apps and APIs that each agent has to be wired into.
- Agents already read and write Markdown well, so we wanted the task list in a format they handle with no extra setup.
- We also wanted a full record of who claimed what and when, without running a database.
Approach
- One file, in Git. Tasks live in TICK.md as plain Markdown with YAML. Every claim, status change and completion is a commit, so the history is built in.
- Claim before you work. An agent claims a task in TICK.md before it starts. A lock file gives other agents an advisory hint about who is writing, but it does not block them. The CLI warns on conflicts, and Git is the source of truth.
- Dependencies that clear themselves. A task can block others. When all of a task's dependencies are done, it unblocks automatically.
- Agents connect through MCP. The tick-mcp-server package gives agents 13 tools to add, claim, comment on, finish and release tasks.
- People get a CLI. The tick command covers init, add, claim, done, list, graph, watch, sync and more.
- Local first. It works offline, your tasks stay in a plain Markdown file in your own repo, and there is no vendor lock-in.
- We used it on its own build. The tick-md repo keeps its own task list in a TICK.md file.
Key Features & Highlights
Core capabilities and innovations that define this project
Plain Markdown tasks: Everything lives in TICK.md, easy for people to read and for agents to parse.
Git history: Every claim, status change and completion is a commit you can review.
MCP server: 13 tools for agents, from tick_add and tick_claim to tick_done and tick_undo.
CLI for people: Install with npm install -g tick-md and run tick init.
Dependency tracking: Tasks unblock automatically once all their dependencies are done.
Filters: Find tasks by status, priority, tag or who claimed them, and export results as JSON.
Dependency graphs: ASCII in the terminal, or Mermaid for your docs.
Results
Open-source license
Tools to install from npm: the tick-md CLI and tick-mcp-server
MCP tools for AI agents
Latest tick-md version on npm, as of Oct 5, 2026
Technologies Used
Want similar outcomes?
Apply for a strategy session or start managed services — we'll match the right tier.
Join The Dispatch
10 AI signals that matter for your business. Every week. No hype.