Summary
Four item types, one search, and files on disk you can still read yourself.
The database is the index and the markdown files in .plan are the human-readable copy, kept in step by the sync tool — so the plan survives being looked at without the server running. Rules being first-class is the unusual part: project conventions live next to the work rather than in a prompt.
What it is
A planning server backed by SQLite with full-text search. It keeps four kinds of item — tasks, features, bugs and rules — and mirrors them to markdown files in a .plan directory inside your project.
What you get
- create_task, update_task, get_task, list_tasks and delete_task, with status, priority and progress
- create_feature, update_feature, get_feature, list_features and delete_feature for user stories and epics
- create_bug, update_bug, get_bug, list_bugs and delete_bug, with severity and reproduction steps
- create_rule, update_rule, get_rule, list_rules and delete_rule for project rules and validations
- search and advanced_search across every plan item, with relevance scoring
- backup_database, restore_database and sync_filesystem
- Resources at plan://index, plan://tasks, plan://features and plan://bugs
Requirements
Node, built from the repository, started against the path of the project you want it to plan. It creates a .plan directory there holding the SQLite database and the mirrored markdown files. No credentials.