AI agents are getting more capable every month. But here is the thing nobody talks about: the harder problem is not making one agent smart. It is getting multiple agents to work together without stepping on each other.

At Purple Horizons, we run AI agents across client projects, internal operations, and product development. Our AI assistant Mia handles everything from email drafts to code reviews to social media scheduling. And as we scaled up, we hit a wall: how do you coordinate multiple agents working on the same codebase, the same project, the same set of tasks?

The answer turned out to be surprisingly simple. A Markdown file.

The Multi-Agent Coordination Problem

If you have been following the AI agent space in 2025 and 2026, you have seen frameworks like CrewAI, AutoGen, and LangChain tackle multi-agent orchestration. These are powerful tools, but they solve a different problem. They focus on how agents communicate within a single workflow or pipeline.

Our challenge was different. We needed agents to coordinate asynchronously across time, across sessions, and across different tools. Think of it like a distributed team where nobody is online at the same time. You need a shared task board that everyone can read, update, and trust.

Most task management tools are built for humans. They require logins, APIs, and complex integrations. AI agents, on the other hand, speak Markdown natively. They can read it, write it, and parse it without any special setup.

Enter tick-md

tick-md is an open-source tool that turns a single TICK.md file into a full multi-agent task coordination system. It is built on Git, designed for natural language interaction, and optimized for workflows where humans and AI agents collaborate.

Here is what makes it different from other AI workflow automation tools:

• No database required. Tasks live in a Markdown file tracked by Git. Every change is a commit with full history.
• File locking. When an agent claims a task, it locks the file to prevent concurrent edit conflicts. No more merge nightmares.
• Dependency tracking. Tasks can block other tasks. When a blocker completes, dependent tasks automatically unblock.
• MCP server. AI agents connect through the Model Context Protocol, making integration seamless with tools like Claude, OpenClaw, and other MCP-compatible platforms.
• Real-time monitoring. A watch mode lets you see task changes as they happen, so you always know what your agents are doing.

How We Actually Use It

Let us walk through a real example. Mia, our AI assistant built on OpenClaw, uses tick-md daily for task tracking across Purple Horizons projects.

A typical workflow looks like this:

1. Gianni creates tasks using the CLI or by editing the TICK.md file directly.
2. Mia claims a task through the MCP server, which locks the file and records the claim in Git.
3. While working, Mia adds comments and progress notes directly to the task.
4. On completion, Mia marks the task done, which automatically unblocks any dependent tasks.
5. Git tracks everything. Every claim, every status change, every completion is a commit with a timestamp and author.

The beauty is that this works whether you have one agent or ten. Each agent reads the same TICK.md file, claims available tasks, and the file locking system prevents collisions.

Battle-Tested by Daily Use

We did not just build tick-md and ship it. Mia has been using it every single day, and that real-world usage surfaced 8 bugs that we fixed in version 1.2.0. That is the advantage of dogfooding your own tools: you find edge cases that unit tests miss.

Some of the issues we caught:

• Race conditions when two agents tried to claim tasks simultaneously
• Edge cases in dependency resolution when tasks had circular references
• File locking behavior on different operating systems
• YAML frontmatter parsing with special characters in task titles

Every bug Mia found made the tool more robust for everyone. That is open source at its best.

Getting Started in 5 Minutes

tick-md is designed to be dead simple to set up:

# Install globally
npm install -g tick-md

# Initialize in your project
cd your-project
tick init

# Add your first task
tick add "Build authentication system" --priority high --tags backend

# Claim it
tick claim TASK-001 @yourname

# Mark it done
tick done TASK-001 @yourname

For AI agents, install the MCP server:

npm install -g tick-mcp-server

Then add it to your MCP configuration and your agents can start coordinating immediately.

Why Markdown Wins

We considered building tick-md on top of a database, or as a SaaS product with a web UI. But Markdown and Git won for three reasons:

1. Transparency. Anyone can open TICK.md and see exactly what is happening. No hidden state, no black boxes.
2. Portability. It works offline, on any machine, with any Git provider. No vendor lock-in.
3. AI-native. Every LLM understands Markdown. You do not need custom parsers or API integrations. Agents can read and write tasks as naturally as they read and write code.

In a world where AI agent coordination is becoming critical infrastructure, simplicity is a feature.

What is Next

tick-md is 100% open source under the MIT license. The protocol, CLI, dashboard, and MCP server are all free forever. We are actively developing new features including a web dashboard for visualization and better support for large-scale agent swarms.

Check out the GitHub repo, try it on your next project, and let us know what you think.

If you are exploring how to bring AI agents into your team workflows, whether for development, operations, or creative work, we would love to chat. At Purple Horizons, we help companies build and deploy AI agent systems that actually work in production. We also offer hands-on training to get your team up to speed.

Want to see how AI agents can transform your team workflows? Book a call with us.