ZenArc MCP

AI-native task management for Claude Code via the Model Context Protocol.

ZenArc replaces scattered TODO.md files with structured, agent-friendly YAML task files. This package provides the MCP server that lets Claude Code create, update, search, and manage tasks across all your projects — directly from the terminal.

Install

From npm

npm install -g zenarc-mcp

From GitHub

npm install -g zenarc-dev/zenarc-mcp

Requires Node.js ≥ 20.

Claude Code Setup

Add to your Claude Code settings (~/.claude/settings.json or a project's .claude/settings.local.json):

{
  "mcpServers": {
    "zenarc": {
      "command": "zenarc-mcp"
    }
  }
}

Restart Claude Code. That's it — no API keys, no ports, no server to run. zenarc-mcp auto-initializes on first use.

First Run

Ask Claude to scan your projects:

"Scan my projects in ~/dev"

Then use natural language for everything:

"List my critical tasks for loa-web"

"Mark the GA4 tracking task as done"

"Create a new high-priority task in codeyourreality to update the hero copy"

"Search for anything related to SEO across all projects"

MCP Tools

| Tool | Description | |------|-------------| | zenarc_scan | Discover projects and add them to the registry | | zenarc_list | List tasks with filters (status, priority, project, tag, assignee) | | zenarc_get | Get full task details by ID | | zenarc_create | Create a new task with structured metadata | | zenarc_update | Update task fields (status, priority, assignee, notes) | | zenarc_search | Keyword search across titles, tags, and notes | | zenarc_context_add | Link file paths or URLs to an existing task |

Task File Format

Each task is a standalone YAML file in {project}/.zenarc/tasks/:

id: tm-20260602-a1b2c3d4
title: Fix GA4 '(not set)' page tracking
status: in_progress
priority: critical
project: loa-web
tags: [analytics, seo, bugfix]
created_at: "2026-05-10T09:00:00Z"
updated_at: "2026-05-18T14:30:00Z"
created_by: human
assigned_to: claude
context:
  files:
    - app/components/PageViewTracker.jsx
    - app/ClientLayout.js
  urls:
    - https://analytics.google.com/analytics/web/
  notes: >
    Fires manual page_view on App Router client-side navigations
    with 100ms delay for document.title to settle.
dependencies: []

Project Registry

Projects are tracked in ~/.zenarc/projects.json:

[
  { "name": "loa-web", "path": "/Users/.../loa/loa-web", "format": "yaml" }
]

The registry is auto-populated during zenarc_scan.

Optional: Firebase Cloud Sync

Set FIREBASE_SYNC_ENABLED=true and GOOGLE_APPLICATION_CREDENTIALS to enable bidirectional sync with Firestore for multi-device access. Without these, ZenArc works fully offline with local YAML files.

Migration from TODO.md

Convert a project's TODO.md into structured YAML tasks:

zenarc-migrate my-project /path/to/my-project

Architecture

zenarc-mcp/
├── src/
│   ├── server.ts          # MCP server (stdio transport)
│   ├── store-init.ts      # Store initialization (local + optional Firebase)
│   ├── core/              # Task schema, YAML store, registry
│   └── sync/              # Optional Firebase Firestore sync layer

Local-first: Tasks are YAML files in each project's .zenarc/tasks/ directory.
Git-native: Tasks version alongside code. Review task changes in PRs.
MCP is local-only: Uses stdio transport. No HTTP/SSE/remote endpoint.

Development

npm install
npm run build    # Compile TypeScript to dist/
npm run dev      # Watch mode
npm start        # Run the MCP server

Tech Stack

TypeScript, Zod (schema validation), YAML (serialization)
@modelcontextprotocol/sdk (stdio transport)
firebase-admin (optional sync layer)

License

MIT

MCP Servers