markplane

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.

Each directory also has an INDEX.md that lists all items with IDs and status — so an AI agent can scan the index (~200 tokens) and load only the specific items it needs, instead of reading everything.

A typical monolithic ROADMAP.md + BACKLOG.md consumes 10,000-30,000 tokens with low signal-to-noise. Markplane's equivalent project state fits in ~1000 tokens.

See the AI Integration Guide for the full context layer design, token budgets, and recommended AI reading patterns.

Installation

Homebrew (macOS and Linux)

brew install zerowand01/markplane/markplane

Shell script (macOS and Linux)

curl -fsSL https://raw.githubusercontent.com/zerowand01/markplane/master/install.sh | sh

Downloads the latest release, verifies the SHA256 checksum, and installs to ~/.local/bin/. Customize with environment variables:

curl -fsSL https://raw.githubusercontent.com/zerowand01/markplane/master/install.sh | INSTALL_DIR=/usr/local/bin sh

Pre-built binary

Download the latest release for your platform from GitHub Releases. Pre-built binaries include the web UI and require no additional dependencies.

macOS / Linux:

tar xzf markplane-v*.tar.gz       # extract the binary from the archive
mv markplane ~/.local/bin/        # or /usr/local/bin/ with sudo

Note: On macOS, binaries downloaded via a browser may be blocked by Gatekeeper. Remove the quarantine attribute with xattr -d com.apple.quarantine ~/.local/bin/markplane, or use the shell script install which avoids this entirely.

Windows: Extract markplane.exe from the .zip to a location on your PATH (e.g., %LOCALAPPDATA%\markplane\).

Build from source

Requires Rust 1.93.0+. To include the web UI, also requires Node.js 18+.

git clone https://github.com/zerowand01/markplane.git
cd markplane

# CLI only
cargo install --path crates/markplane-cli

# CLI + Web UI (single binary)
cd crates/markplane-web/ui && npm install && npm run build && cd ../../..
cargo install --path crates/markplane-cli --features embed-ui

Initialization

Initialize Markplane in your project:

markplane init --name "My Project"

This creates the .markplane/ directory with config, templates, and starter content to explore. Use markplane init --empty to skip starter content.

.markplane/
├── config.yaml           # Project settings
├── INDEX.md              # Root navigation
├── roadmap/              # Epics — strategic goals and phases (EPIC-xxxxx)
├── backlog/              # Tasks — the "what" to do (TASK-xxxxx)
├── plans/                # Plans — the "how" to do it (PLAN-xxxxx)
├── notes/                # Notes — research, ideas, decisions (NOTE-xxxxx)
├── templates/            # Document templates
└── .context/             # Token-optimized summaries for AI

Web UI

Launch the local web dashboard to see your project:

markplane serve --open

This opens http://localhost:4200 in your browser with a dashboard, kanban board, dependency graph, and more.

See the Web UI Guide for keyboard shortcuts, views, and details.

AI Integration

Markplane includes a built-in MCP server that lets AI coding assistants manage your project directly. Connect your assistant, then just tell it what to do in natural language — "create a task for the login bug", "show me what's blocked", "mark TASK-fq2x8 as done".

Example with Claude Code:

Per-user — adds Markplane to your local editor configuration:

claude mcp add --transport stdio markplane -- markplane mcp

Project-wide — add a .mcp.json file at the repo root so every team member gets the integration automatically:

{
  "mcpServers": {
    "markplane": {
      "command": "markplane",
      "args": ["mcp"]
    }
  }
}

See the MCP Setup Guide for other editors, configuration options, and the full tool and resource catalog.

CLI

The CLI is a power-user interface for everything Markplane can do:

markplane add "Fix login redirect" --type bug --priority critical --tags auth
markplane ls                          # List tasks
markplane ls --priority high,critical # Filter by priority
markplane show TASK-fq2x8            # View item details
markplane start TASK-fq2x8           # Set to in-progress
markplane done TASK-fq2x8            # Mark as done
markplane dashboard                   # Project overview
markplane sync                        # Regenerate INDEX.md + .context/
markplane check                       # Validate cross-references

See the CLI Reference for complete command documentation.

Status Workflows

Task statuses are fully configurable. Each status maps to one of six status categories (draft, backlog, planned, active, completed, cancelled) that control system behavior (kanban columns, progress