doc-search

Indexandria — Documentation Fetcher

You have access to Indexandria's MCP tools for pulling web documentation into the conversation. Nothing is stored on disk — all content lives in memory for the session.

Two Workflows

Quick Mode — crawl_docs

Best for 1-15 pages. Content goes directly into context.

crawl_docs("https://react.dev/reference/react/useEffect", depth=1)

Use when:

• User provides a specific docs URL
• You need a single API reference page
• User says "check the docs" for a specific topic

Deep Mode — index_docs + get_indexed_page + search_indexed

Best for 20-100 pages. Crawls everything into memory, returns a compact index. You then selectively retrieve only the pages you need.

# Step 1: Index the docs (returns a table of contents, not full content)
index_docs("https://fastapi.tiangolo.com/tutorial/", depth=2, max_pages=50)

# Step 2: Search for what you need
search_indexed("OAuth2")

# Step 3: Read the relevant pages
get_indexed_page([12, 15])

Use when:

• Working with a full framework or library
• User asks to "learn" or "study" documentation
• Multiple related topics need to be referenced
• A large reference site needs targeted navigation

Choosing the Right Mode

Scenario	Mode	Example
Single API/hook/function	Quick, depth=1	crawl_docs("https://react.dev/reference/react/useEffect")
One tutorial page	Quick, depth=1	crawl_docs("https://fastapi.tiangolo.com/tutorial/first-steps/", depth=1)
Tutorial section with subpages	Quick, depth=2	crawl_docs("https://docs.python.org/3/tutorial/classes.html")
Full framework docs	Deep	index_docs("https://fastapi.tiangolo.com/tutorial/", max_pages=60)
Large API reference	Deep + patterns	index_docs("https://react.dev/reference/", include_patterns=["*/reference/*"])
Unknown docs structure	Deep, low pages	index_docs(url, max_pages=30) then browse the index

Using Docs While Writing Code

When you have indexed documentation, follow this pattern before writing implementation code:

1. Search before implementing. Before writing code that uses a library API, call search_indexed for the specific function, class, or pattern you need. Do not rely on memory — the docs are right there.

2. Read the relevant pages. Use get_indexed_page to pull in the full content of pages that match your search. Read them carefully.

3. Verify parameters and return types. If the docs specify exact parameter names, types, or return values, use them as written. Do not guess.

4. Re-search when unsure. If mid-implementation you are unsure about a behavior, call search_indexed again rather than making assumptions.

Tool Reference

crawl_docs(url, depth=2, max_pages=15, include_patterns=None, exclude_patterns=None)

Crawl and return content directly. Output capped at ~110 KB.

index_docs(url, depth=2, max_pages=50, include_patterns=None, exclude_patterns=None)

Crawl up to 100 pages into memory. Returns a compact index table (page number, title, headings). Pages stay in memory until the next index_docs call or session end.

get_indexed_page(page_numbers)

Retrieve full content of specific pages by number (1-based). Example: get_indexed_page([1, 5, 12]). Output capped at ~110 KB — request fewer pages if hitting the limit.

search_indexed(query, max_results=10)

Case-insensitive keyword search across all indexed pages. Returns matching snippets with page numbers and context. Use results to decide which pages to read fully.

Tips

• Be specific with URLs — point to the relevant section, not the site root.
• Use include_patterns to stay within a docs section: ["*/reference/*"]
• Use depth=1 when you only need a single page.
• Deep index first, read selectively — for large sites, index everything but only read what you need.
• Re-index when switching topics — index_docs replaces the previous index.

Example: Full Workflow

User: "Build a FastAPI app with OAuth2 and JWT authentication"

1. index_docs("https://fastapi.tiangolo.com/tutorial/", depth=2, max_pages=60)
   → Returns index: 45 pages covering tutorial sections

2. search_indexed("OAuth2")
   → Matches: Page 23 (Security - OAuth2), Page 24 (OAuth2 with JWT)

3. get_indexed_page([23, 24])
   → Full content of OAuth2 tutorial pages

4. search_indexed("dependency injection")
   → Matches: Page 8 (Dependencies), Page 9 (Dependencies in path operations)

5. get_indexed_page([8, 9])
   → Full content of dependency injection pages

6. Write the implementation using the docs as reference