Projects

Identity is the git remote, not the path

A project's identity is its git remote URL, falling back to the common git directory when there's no remote. This is deliberate:

• It unifies worktrees -- the same repository checked out in multiple worktrees shares one backlog instead of fragmenting it per checkout.
• It survives re-cloning to a different path.
• It survives renames.

The MCP server needs zero configuration

magpie serve-mcp derives its project from its own spawn directory by shelling out to git remote get-url origin (and git rev-parse --git-common-dir as the fallback). An agent host (Claude Code, Cursor) launches the server from wherever it's already working, so the right project is scoped automatically -- nothing to configure per-repository.

Optional branch constraint

A capture can carry an optional branch value. When set, only an MCP session on that exact branch can lease it with queue_take -- useful for work that's only valid against a specific branch or worktree. NULL (the default) means any session in the project can take it.

Projects are not tags

Project scoping uses a dedicated project_id column, not a tag. A typo in a tag name would silently mis-scope agent work with no error; a project derived from git remote get-url can't be typo'd. Tags remain for cross-cutting themes (#bug, #refactor) that intentionally span projects.