AI reference

https://dia.rakudeji.com/mcp

dia stores node-and-edge diagrams grouped into documents; documents are grouped into projects.
Every document belongs to a project; the project "default" exists from the start.
Create a document with create_document, then add diagrams with write_diagram.
Find existing projects and documents with list_projects.
Inspect a document with read_document. Reorder or retitle with write_document by
rearranging the diagrams array; there are no positional indexes.
A diagram is a single JSON object with title, description, model (structure),
style (appearance), and optional presets (named style values shared between nodes).
Edits are sent as JSON Patch; never resend the whole diagram.
Coordinates and sizes are computed automatically and cannot be specified.
Nodes are created from ids used in edges, so model.nodes is usually unnecessary.
Every write appends a version. See list_versions and restore_diagram.

Readable diagrams:
- One diagram answers one question. Split a large system into several diagrams in one document
  (an overview, then one diagram per area or flow), or add views that highlight each flow.
- Draw one kind of relation as arrows in a diagram (data flow, calls, or control);
  state the other kinds in labels or the description instead of adding crossing arrows.
- Keep labels to a few words; explanations, assumptions, and numbers go in description.
- To build a picture up step by step, or to compare before and after, use a "steps" document.
- Check each diagram with render_diagram and revise what reads poorly.

For table-like diagrams, preset the container { gap 0, padding 0, paddingY 12, flow right }
and the rows { padding 16, paddingY 4, minHeight 0 }.