doctree-mcp: Skill for Claude Code

Retrieve Documentation

Find and retrieve content from the markdown knowledge base using the doctree-mcp tools.

Workflow

Follow these steps in order. Do NOT skip the tree step — it lets you target exactly what to read instead of dumping entire documents.

Step 1: Search or browse

If you have a specific query, search for it:

search_documents(query: "$ARGUMENTS")

If you're exploring broadly or want to see what's available:

list_documents(query: "$ARGUMENTS")

Note the doc_id and node_id values from the results.

Step 2: Read the document outline

Pick the most relevant document from step 1 and get its structure:

get_tree(doc_id: "<doc_id from step 1>")

Read the outline carefully. Identify which sections are relevant to the query based on their titles, word counts, and summaries. Do NOT retrieve everything — pick only what matters.

Step 3: Retrieve specific content

For a single section or a few sections:

get_node_content(doc_id: "<doc_id>", node_ids: ["<node_id>", ...])

For a section and all its subsections (more efficient than fetching each child):

navigate_tree(doc_id: "<doc_id>", node_id: "<parent_node_id>")

Step 4: Follow cross-references

If the retrieved content mentions or links to other documents, repeat steps 2-3 for those documents if the user needs that information.

Tips

• Use specific terms in search queries — BM25 matches keywords, not semantic meaning
• Use facet filters to narrow search: filters: { "type": "runbook", "tags": ["auth"] }
• Check search snippets first — the top 3 results include inline content, which may already answer the question
• Prefer navigate_tree over multiple get_node_content calls when you need a whole section with children
• If search returns no results, try alternative terms or check the glossary — abbreviations like "CLI" expand to "command line interface" automatically