indexandria

Indexandria

Crawl any web documentation and pull it straight into Claude Code's context.

Indexandria is a Claude Code plugin that fetches documentation from any URL, converts it to clean markdown, and returns it directly into the conversation. Claude then uses this content as context while writing code — no database, no background server, no disk writes.

Named after the Great Library of Alexandria.

Install

/plugin marketplace add alicankiraz1/indexandria
/plugin install indexandria@indexandria

Requires Python 3.10+ and uv.

Quick Start

Tell Claude to fetch some docs:

> Crawl https://fastapi.tiangolo.com/tutorial/ and then build me an API with validation

For larger documentation sites, use deep indexing:

> Index the full React reference docs, then help me build a custom hook

Two Workflows

Quick Mode — Small, Focused Crawls

crawl_docs fetches 1-15 pages and returns content directly into context. Best for single API lookups and short tutorials.

"Crawl the useEffect docs"
       │
       ▼
 ┌───────────┐     ┌────────────┐     ┌───────────┐
 │  Claude    │────▶│ crawl_docs │────▶│  Crawler   │
 │  Code      │     │ (MCP tool) │     │ httpx+BS4  │
 └───────────┘     └────────────┘     └───────────┘
       │                                     │
       │          markdown content           │
       │◀────────────────────────────────────┘
       ▼
 Claude writes code using the docs

Deep Mode — Large Documentation Sites

index_docs crawls up to 100 pages into memory and returns a compact table of contents. Claude then uses search_indexed and get_indexed_page to selectively read only what's needed.

"Index the FastAPI tutorial, then build OAuth2 auth"
       │
       ▼
 ┌───────────┐     ┌────────────┐     ┌───────────┐
 │  Claude    │────▶│ index_docs │────▶│  Crawler   │
 │  Code      │     │            │     │ (100 pages)│
 └───────────┘     └────────────┘     └───────────┘
       │                                     │
       │          compact index (TOC)        │
       │◀────────────────────────────────────┘
       │
       ├── search_indexed("OAuth2") ──▶ page 23, 24
       │
       ├── get_indexed_page([23, 24]) ──▶ full content
       │
       ▼
 Claude writes code with precise docs reference

No database. No files. No background process. Everything lives in memory for the session.

Tools

crawl_docs — Quick mode

Parameter	Default	Description
url	required	Starting URL to crawl
depth	2	Link levels to follow (1-3)
max_pages	15	Maximum pages to fetch (up to 50)
include_patterns	none	URL globs to include (e.g. ["*/docs/*"])
exclude_patterns	none	URL globs to exclude (e.g. ["*/blog/*"])

index_docs — Deep mode

Parameter	Default	Description
url	required	Starting URL to crawl
depth	2	Link levels to follow (1-3)
max_pages	50	Maximum pages to index (up to 100)
include_patterns	none	URL globs to include
exclude_patterns	none	URL globs to exclude

get_indexed_page — Read from index

Parameter	Default	Description
page_numbers	required	List of 1-based page numbers (e.g. [1, 5, 12])

search_indexed — Search across index

Parameter	Default	Description
query	required	Search term (case-insensitive)
max_results	10	Maximum matches to return

Output from all tools is capped at ~110 KB to stay within Claude Code's MCP token limit.

Tips

• Be specific — point to the relevant docs section, not the entire site
• Quick mode for 1-5 pages, deep mode for 20+ pages
• Use include_patterns to stay within a docs section: ["*/reference/*"]
• Search before reading — with deep mode, search first to find the right pages
• Only http and https URLs are accepted

Security & Privacy

• No data stored — nothing is written to disk, ever
• No external services — all processing is local
• URL validation — only http/https schemes are allowed
• Session-scoped — indexed content exists only in memory, cleared when the session ends
• Open source — the entire codebase is ~500 lines of Python

Contributing

github.com/alicankiraz1/indexandria

License

MIT