gutenberg-mcp

An MCP server that gives an LLM direct search and read access to Project Gutenberg's catalog of 75,000+ free public-domain books.

Tools

Tool	Purpose
search_books(query, author)	Search the catalog by title and/or author
search_books_by_subject(subject)	Search by subject/topic
get_book_info(book_id)	Metadata + approximate word count for a book
get_book_text(book_id, start_word, word_count)	Fetch a word-range slice of a book's plain text

get_book_text is paginated by word count (default 3,000 words, ~10-12 pages) rather than returning entire books at once, so an LLM can pull exactly the section it needs without blowing its context window.

Why this exists (and not the gutendex API)

Most Gutenberg MCP servers wrap the third-party gutendex.com API. As of mid-2026 that service sits behind a Cloudflare JS challenge that a plain HTTP client can't solve — every request gets a 403, regardless of headers or user-agent. Several existing "Gutenberg MCP" repos on GitHub are broken for this same reason.

This server goes straight to Project Gutenberg instead:

• Search is built from Gutenberg's own sanctioned bulk catalog feed (gutenberg.org/cache/epub/feeds/pg_catalog.csv) — the same data Gutendex itself is built from. Gutenberg's robots.txt explicitly disallows crawling /ebooks/search, so this server doesn't scrape that page; the bulk feed is the sanctioned way to get catalog data for search.
• Book text is fetched directly from Gutenberg's own plaintext mirrors (gutenberg.org/cache/epub/{id}/pg{id}.txt), which robots.txt does allow.

The catalog CSV (~21MB) is fetched once per server process and cached in memory — not re-downloaded per call.

Running it

docker build -t gutenberg-mcp-server .

Add to your MCP client config (e.g. Claude Desktop / Claude Code):

{
  "mcpServers": {
    "gutenberg": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "gutenberg-mcp-server"]
    }
  }
}

License

MIT