lemon-manuals-mcp
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
lemon-manuals-mcp
Prototype MCP server (streamable HTTP, stateless) for the self-hosted
LEMON/CHARM car-repair-manual archive (home-lab repo
kubernetes/apps/lemon-manuals/).
Tools:
list_makes— all makes with vehicle counts and year rangessearch_vehicles— free-text over make/model/engine/year → grouped drivetrains, variants, and all source manualsget_page— opaque encoded site path → normalized markdown, with tree depth and byte controlssearch_manual— persistent FTS5 search over titles and/or body text, deduplicated by document contentget_image— opaque encoded image path → validated MCP image block (2 MB maximum; never truncated)
All returned paths are opaque encoded tokens. Pass them to get_page,
search_manual, or get_image exactly as returned; never decode or rebuild them.
search_manual defaults to mode: "both"; title and body are also
available. Results contain a body-text snippet, an also_under list for
duplicate tree placements, and parsed from/through applicability when the
publisher labels it. Image-only pages are identified explicitly.
Browse/search is answered locally from the databases' index.json files
(~305k vehicles, ~200ms load); page content is fetched from the lemon-website
server and converted with turndown. Relative and /hyperlink/... references are
normalized to absolute encoded site paths that can be passed directly back to
the tools.
Manual full-text data lives in data/manual-search.sqlite by default. A first
search tree-walks and persists an unindexed manual. mode: "title" only fetches
matching leaves; body/both completes the manual's body index and can
therefore be slow on first use. Production should prebuild the index:
# One or more opaque vehicle roots (recommended for incremental rollout)
bun run index-manuals -- '/Volkswagen/2005/Touareg%20%287LA%29%20V8-4.2L%20%28BHX%29/'
# No arguments indexes every isComplete vehicle. This is a full-corpus job.
bun run index-manualsOCR evaluation
The bounded OCR pilot report records CPU/GPU/VLM comparisons on DGX Spark and proposes a portable archive, enrichment and retrieval architecture. It includes methodology and reproduction limits, but no manual-derived images or transcripts. OCR integration and the proposed deployment options are not implemented yet.
The fresh-image source-span evaluation compares crop-boundary safeguards and local orientation with the frozen OCR baseline, keeping candidate recall, provisional selection and human acceptance separate.
Related MCP server: docs-mcp
Setup
The index files are gitignored (147MB, re-fetchable from the cluster):
kubectl -n lemon-manuals exec deploy/lemon-website -- cat /data/charm/index.json > data/charm-index.json
kubectl -n lemon-manuals exec deploy/lemon-website -- cat /data/lemon/index.json > data/lemon-index.jsonThe server needs HTTP reach to lemon-website. For local dev:
kubectl -n lemon-manuals port-forward deploy/lemon-website 18080:8080Env: LEMON_BASE_URL (default http://127.0.0.1:18080), LEMON_DATA_DIR
(default ./data), and LEMON_SEARCH_DB (default
$LEMON_DATA_DIR/manual-search.sqlite). The search database directory must be
writable and persistent if lazy indexing is enabled.
Run / register
With Bun available on PATH, either installed directly or through an activated
mise environment:
bun src/index.ts # HTTP server on :8787 (/mcp, /healthz)
bun test/client.ts # end-to-end smoke (needs port-forward)
bun run live-corpus # one complete manual/corpus/decade
bun test # hermetic regression suite
bun run typecheck
claude mcp add --transport http lemon-manuals http://127.0.0.1:8787/mcpThe pages.mtbl direct-read option (vendored oxidized-mtbl crate in
kelchm/lemon-website) can eventually replace the HTML ingestion round-trip.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Search and read Apple developer documentation, HIG, and WWDC transcripts as Markdown.
Read web pages as markdown, search the web, list a site's URLs and run a real Chrome.
Web search, URL content extraction to Markdown, site mapping, and recursive web crawler.
Live web search results and web pages as Markdown; page fetches use a residential network.
Related MCP Servers
- AlicenseAqualityBmaintenanceMCP server for markdown files — search, extract sections, list headings, find code blocks across docs.678 npm8MIT
- FlicenseNot gradedqualityDmaintenanceIndexes documentation sites by base URL and serves keyword search, optional semantic search, and Markdown page retrieval as MCP tools, all from a single SQLite file.-
- AlicenseAqualityDmaintenanceEnables searching, reading, and listing pages from the official chezmoi documentation. Supports caching and Markdown conversion for LLM-friendly access.4MIT

docs-mcpofficial
AlicenseAqualityBmaintenanceLets AI agents search and read a folder of Markdown documentation at query time, exposing heading-level search, whole-page or section retrieval, and page listing as tools and resources over stdio or Streamable HTTP. Can run as a CLI, a hosted endpoint, or a library inside an existing Node or Next.js site.3206 npm1MIT