E

Planure: MCP Content Pagination (fix truncation on large pages)

Status: Implemented. The authoritative planure lives at feature_Planure/feature_Planure_mcp_content_pagination.md; this page mirrors it for the documentation site.

Problem

get_document returned raw markdown and rendered HTML in a single response. Large guides produced several-hundred-kilobyte tool results that MCP hosts truncated, so agents could not read them in one call.

Change

  • get_document(path, format="markdown", offset=0, max_chars=0) — one representation (markdown default, html opt-in), paginated with truncated / next_offset.
  • get_document_meta(path) — metadata, headings, neighbours, and sizes without the body.
  • get_document_section(path, heading, ...) — one markdown section by heading id or text.
  • MCP_MAX_CONTENT_CHARS setting (default 40000) bounds characters per call.
  • The docs://<path> resource is unchanged (already markdown-only).

Files

  • apps/documentation/mcp_server/server.py
  • config/settings.py
  • docs/Developer/mcp-documentation-server.md

Validation (user-run — agent does not test)

  1. manage.py run_mcp.
  2. get_document_meta on a large guide → metadata only, no body.
  3. get_document with no arguments → markdown only, no html, truncated=false under the cap.
  4. Oversized page → follow next_offset until truncated=false; slices reconstruct the body.
  5. get_document_section by heading id and by text → only that section.
  6. format="html" returns HTML; an unknown format raises.

Notes

  • Breaking tool-shape change: html is no longer in the default get_document response.
  • Section slugs are best-effort; heading-text matching is the fallback.