Source of Truth
Note. Server-side Source of Truth indexing and the MCP graph tools (
get_doc_graph,read_doc_sections,reindex_doc_graph, and the 17doc_*tools) were removed in v0.22.0. The graph now lives entirely on the agent's machine: there is no hosted index, no reindex quota, and nothing about it draws on your project balance.
How do I give an agent the Source of Truth graph?#
Run markdown-lsp in the repository you want the agent to query, and the graph is available to it for as long as it works there. Nothing is uploaded and nothing needs to be enabled in Docsbook.
The graph search is powered by markdown-lsp — our open-source LSP implementation for Markdown, published on npm as markdown-lsp. Run it in any repo whose docs you want to query:
npx markdown-lsp <subcommand> ./docsThis makes the LSP-style tools available to the agent against your local working tree. See the markdown-lsp README for the full subcommand list and setup options.
What the graph contains#
For each page the graph stores:
- Canonical reference (
path#section) - Title and frontmatter
- Heading tree with stable anchors
- Section bodies (markdown)
- Outbound and inbound links
The graph is rebuilt incrementally from disk on every relevant edit, so it always matches the working tree the agent sees.
How the graph is built#
The graph is parsed by markdown-lsp — our open-source Language Server Protocol implementation for Markdown, published on npm as markdown-lsp. It uses unified + remark AST instead of regex, so:
- Relative paths like
../guide.md#sectionresolve correctly - Wiki-style links
[[note]]are supported - Broken links are detected up-front and surfaced as
unresolveditems - Inline, reference, and autolink styles are handled uniformly
LSP-style tools available to the agent#
These are MCP tools the plugin exposes to Claude Code (or any MCP-compatible client). They run against the in-memory graph, so they are free and instant — no API calls back to Docsbook.
Structure
| Tool | What it returns |
|---|---|
doc_outline |
Heading tree of one page, no bodies |
doc_workspace_outline |
All pages with titles and heading counts |
doc_get_section |
Body of a specific section by anchor |
Search
| Tool | What it returns |
|---|---|
doc_search_symbols |
Fuzzy subsequence over headings; query oaf matches OAuth flow |
doc_search_text |
Full-text search with surrounding-context snippets + line/col |
doc_search_paths |
Glob filter on page paths (docs/*.md, **/auth.md) |
Link graph
| Tool | What it returns |
|---|---|
doc_search_links_to |
All incoming links to a page (LSP references) |
doc_search_links_from |
All outgoing links from a page |
doc_resolve_link |
Convert a relative or wiki link into a resolved target page + anchor |
Why is the graph local instead of hosted?#
Docsbook builds the Source of Truth graph on the agent's machine because the three things an agent needs from it — freshness, privacy and unlimited re-reads — are exactly the three a hosted index cannot give.
- No quotas, and no cost. Reindex as often as the agent needs; it is all on disk, and no call is metered.
- Always fresh. The graph reflects uncommitted edits the moment the agent saves them, which a hosted index built from pushed commits cannot do.
- Private. Unpublished drafts never leave the machine.
- Not tied to Docsbook.
markdown-lspruns against any Markdown repository, including documentation that is not published here.
The trade-off is real and worth naming: an agent with no checkout of your repository gets nothing from this graph. That agent should use the hosted search_docs and get_doc_outline tools instead.
Related#
- MCP Server — the hosted server for workspace, content, analytics and webhooks.
- Docs Skills — the skills catalog that builds on top of the graph.
- llms.txt — the machine-readable index of the published site, for agents with no checkout.
- Webhooks — subscribe to
content.indexedandcontent.outdatedon the hosted side.