mempalace-read-docs

Read documentation from MemPalace

This skill recalls documentation that was filed into MemPalace (see the mempalace-ingest-docs skill). It supports three jobs:

1. Answer a question — semantic search across the docs.
2. Browse — list what is stored (wings → rooms → drawers).
3. Export — reassemble a whole document.

Access

You can reach the server two ways — use whichever is available:

1. MCP tools (mempalace_*) — preferred when connected.
2. REST API (/mp/api/v1) via curl — when ENABLE_REST_API=true.

A read-only API key is enough for everything in this skill.

Job 1 — Answer a question

1. Run a semantic search with the user's question.
2. Read the top results, then answer from them. Quote or cite the source_file / wing / room so the user can verify.
3. If results are weak, broaden the query or raise the limit, and say so.

mempalace_search({ "query": "how do I rotate an API key?", "limit": 5 })

REST:

KEY="YOUR_READONLY_KEY"; BASE="http://localhost:8000/mp/api/v1"
curl -sS -X POST "$BASE/search" \
  -H "Authorization: Bearer $KEY" -H "Content-Type: application/json" \
  -d '{"query": "how do I rotate an API key?", "limit": 5}'

To stay inside one document, pass "wing": "acme-api-docs" in the search.

Job 2 — Browse

• mempalace_list_wings → which documents exist.
• mempalace_list_rooms({ "wing": "acme-api-docs" }) → its sections.
• mempalace_list_drawers({ "wing": "...", "room": "..." }) → the chunks.
• mempalace_get_drawer({ "drawer_id": "..." }) → one chunk in full.

REST equivalents: GET /wings, GET /rooms?wing=, GET /drawers?wing=&room=, GET /drawers/{id}.

Job 3 — Export a whole document

1. list_rooms for the wing to get every section.
2. For each room, list_drawers to get its chunks (use limit/offset to page through all of them — do not stop at the first page).
3. Concatenate the drawer contents, grouped by room, in order. The breadcrumb line at the top of each drawer tells you where it belongs.
4. Output clean Markdown.