exocortex-project

A personal project-management layer on a self-hosted fork of Plane Community Edition, for a solo developer who works with Claude Code and with OpenAI Codex run by Symphony. The loop:

capture ──► clarify ──► dispatch ──► Human Review
 note        capture     next         review

You say "note: ..." and the agent records your words verbatim. A clarify pass turns notes and ideas into well-specified work items with parent and blocked-by links. The ready queue picks what can start now, and the item is worked in your session or handed to Codex through Symphony. Finished work waits in Human Review with its patch until you approve it or send it back with a comment. The agent does all of this through MCP tools. You use the Plane web app and talk to the agent.

Status (0.1.0): setup --bare and the full loop with a stand-in Codex have been tested. plane up --docker / setup --docker and Symphony runs with real Codex are not yet verified.

What the wheel contains

• Client core (exocortex_project.core): a client for Plane's /api/v1/, the ready rule over blocked_by, context packs for one item, capture with parent and blocker links, and the Human Review queue with per-file patch stats.
• MCP server exocortex-project-mcp (stdio): every everyday operation as a tool, the procedures as prompts, the state vocabulary and the working rules as server instructions.
• One-time commands exocortex-project: guide, version, setup, init, doctor, plane up|down|status|bootstrap|backup|restore, symphony workflow|check.
• The Plane fork server: the Django API with Celery worker and beat, its migrations, a bootstrap_exocortex management command, and the prebuilt web app, served by the API on the same port (no Node, no proxy). The public boards, God Mode admin and realtime servers are not included.
• Offline docs: SETUP.md, USAGE.md and ARCHITECTURE.md.

Not bundled: PostgreSQL, Redis or Valkey, and MinIO or S3 (for --bare), or Docker (for --docker); Symphony with a Plane tracker adapter, and the Codex CLI (optional); Claude Code or another MCP client.

Install and first run

Python 3.11 or newer.

pip install exocortex-project        # or: pipx install exocortex-project

Then, in the repository the work is about, tell your agent:

exocortex-project is installed, set it up

The agent reads the setup checklist with exocortex-project guide and runs it. To do it yourself:

exocortex-project guide                                    # the full SETUP.md
exocortex-project setup --bare --workspace <slug> --project <IDENT>
# or: exocortex-project setup --docker --workspace <slug> --project <IDENT>

setup runs guide, plane up, plane bootstrap, init and doctor in order and stops at the first failure, saying how to fix it. Every step is safe to repeat. Afterwards, log in to the web app with the admin account that plane bootstrap printed, change its password, and approve the exocortex-project MCP server when your client asks.

MCP tools and prompts

17 tools:

4 prompts: note, capture (also the clarify pass over 待整理), next and review. In Claude Code they appear as /mcp__exocortex-project__<name>. For clients without MCP prompts, init writes the same procedures as the skills pm-note, pm-capture, pm-next and pm-review.

Configuration

The package stores no data of its own. Work items live in Plane's Postgres and storage bucket, and plane backup / plane restore cover both.

More

exocortex-project guide prints the full SETUP.md: prerequisites, every step with its expected output, all doctor checks, the optional Symphony setup, and upgrades, backups and rollback. USAGE.md and ARCHITECTURE.md are next to it in the installed package.

License

AGPL-3.0-only, like the Plane code it includes. This project is not affiliated with or endorsed by Plane Software, Inc.