mempalace-ingest-docs

Ingest documentation into MemPalace

This skill loads a long document into a MemPalace server so it can be recalled later by meaning. A document is broken into small pieces and filed using the palace structure:

• Wing — the whole document or product (e.g. acme-api-docs)
• Room — a top-level section (e.g. authentication, getting-started)
• Drawer — one chunk of content (a subsection or a few paragraphs)

Before you start

You need a running MemPalace server and a write API key. You can reach it two ways — use whichever is available:

1. MCP tools (mempalace_*) — if the MemPalace MCP server is connected to this client. Preferred.
2. REST API (/mp/api/v1) via curl — if ENABLE_REST_API=true on the server. Use the same write key.

If neither the wing/room are given, ask the user for the document and a short wing name (a slug for the doc/product).

Steps

1. Read the document. Load the file the user points to (or the text they paste).

2. Plan the taxonomy.

   • Pick one wing = a slug for the whole document (e.g. acme-api-docs).
   • Use each top-level heading (# / ##) as a room (slug it, e.g. "Getting Started" → getting-started).
   • Keep the original heading path at the top of each chunk so it stays readable on its own.

3. Chunk the content. Split each section into drawers of roughly 200–400 words. Never split in the middle of a code block or table. Each drawer must make sense on its own — prepend a one-line breadcrumb such as # Acme API > Authentication > API keys to the content.

4. Avoid duplicates. For each chunk, call mempalace_check_duplicate (or skip if re-ingesting on purpose). Skip chunks that already exist.

5. File each chunk with mempalace_add_drawer:

   • wing = the document slug
   • room = the section slug
   • content = the breadcrumb + chunk text
   • source_file = the original file path
   • added_by = docs-ingest

6. Link related sections (optional). If two sections in different wings are related, create a mempalace_create_tunnel between their rooms.