Markplane

Project management in your repo, built for AI collaboration.

---

Every developer using AI coding assistants hits the same wall: the AI is great at code, but clueless about the project. It doesn't know what you're building, what's blocked, or what's next. Meanwhile, your project data is locked in a SaaS tool that lives outside your codebase, outside your editor, outside your flow.

Markplane stores every task, epic, and plan as a markdown file inside your repo — version-controlled with git, browsable with any editor, and automatically compressed into token-efficient summaries your AI assistant can read and act on. No database, no SaaS, no context-switching.

Quickstart

1. Install markplane (see below)
2. Initialize in your project: markplane init --name "My Project" (details)
3. Open the web UI: markplane serve --open
4. Connect your AI via MCP (setup guide)
5. Just tell your AI what to do in plain English: "Create a task for the login bug, mark it critical, and link it to the auth epic."

A day with Markplane

Morning — You open your AI assistant and ask "what should we work on?" It pulls the project summary via MCP — 2 tasks in progress, 1 blocked on the auth migration, the payments epic is 60% done — and suggests the highest-priority planned task.

Mid-day — You hit a design question. You talk it through with your AI, arrive at an approach, and say "create a task for this, and then let's work on an implementation plan." Markdown files appear in your repo, linked and ready.

Afternoon — "Let's implement TASK-xyz." Your AI reads the plan, understands the dependencies and project context, and builds it. When you're happy with the result: "mark it done." The context layer updates.

End of day — You commit. Your project state is versioned alongside your code. Tomorrow's session picks up where today left off — no re-explaining.

Why Markplane

• AI-native, not AI-retrofitted — .context/ summaries compress full project state into ~1000 tokens. Your AI loads only what it needs. Every design decision optimizes for LLM context windows.
• Files are the database — No vendor, no subscription, no migration. grep your tasks. git blame your status changes. Branch your backlog like you branch your code.
• MCP server built in — AI assistants don't just read your project — they manage it. Query tasks, update status, create plans, and track dependencies without leaving your conversation.
• Zero infrastructure — markplane init and you're done. No signup, no server, no Docker container. It's a single binary.

Agent Memory

Markplane also works as structured memory for autonomous AI agents. Instead of unstructured daily logs that degrade over time, your agent gets typed tasks, decisions, and project state — compressed into a token-efficient summary that persists across sessions.

For OpenClaw, see @zerowand/markplane-memory.

The Context Layer

Markplane automatically generates a .context/ directory with compressed, token-efficient summaries of your entire project state. This is what makes AI collaboration work — your assistant doesn't need to read every file to understand the project.