riida-mcp

MCP server for riida — infer and update book metadata by reading PDF content and file paths.

Requirements

• Node.js 22 or later
• riida installed and run at least once (creates the database)
• (optional) poppler's pdftotext on PATH — improves read_pdf_colophon ISBN coverage for PDFs whose colophon fonts pdf.js cannot read (brew install poppler / apt install poppler-utils)

Setup

Add to your Claude Code project config (.mcp.json):

{
  "mcpServers": {
    "riida": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "riida-mcp"]
    }
  }
}

Or for Claude Desktop, add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "riida": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "riida-mcp"]
    }
  }
}

Tools

search_books

Search the library by any combination of filters.

list_books_needing_metadata

Lists books that are missing title or authors. Returns file paths with directory structure useful for metadata inference.

get_book_metadata

Gets the current stored metadata for a book.

read_pdf_pages

Extracts plain text from the first N pages of a PDF file. Use this to infer title, authors, and publisher from the book content.

read_pdf_colophon

Extracts the colophon (奥付) from the last pages of a PDF and parses its bibliographic data. This is the counterpart to read_pdf_pages: Japanese books print the ISBN, first-edition date, and publisher on a final colophon page.

The result includes total_pages, the raw tail text, an isbn_source (pdfjs, pdftotext, or null — see below), and a parsed colophon object:

The own-book ISBN is chosen over advertised titles in the back matter by its Japanese C-code and its proximity to colophon keywords (発行所 / Printed in Japan). isbn_confidence is low when several ISBNs were found but none could be confirmed as the book's own — typically a back-cover ad list when the real colophon is image-only and not text-extractable. Always cross-check a low result (and the release_date / publisher, which are best-effort) against the returned raw text. ISBNs printed only as a barcode image cannot be extracted.

pdftotext fallback. Some colophons use fonts without a ToUnicode CMap, so pdf.js (the default parser) extracts little or no usable text from them while poppler's pdftotext reads them cleanly. O'Reilly Japan is the common case: its colophon ISBN and "Printed in Japan" are ASCII and survive, but the Japanese 発行日 / 発行所 do not — so pdf.js reports a confident ISBN yet an empty date and publisher. The tool therefore re-reads the trailing pages with pdftotext whenever any of {ISBN, release_date, publisher} is missing, then takes from poppler only the fields pdf.js could not read. If poppler supplies a strictly better ISBN, isbn_source reports pdftotext; otherwise the pdf.js ISBN is kept and only the date/publisher are filled. pdftotext is run with -layout so multi-column colophons keep each date beside its 第N版第N刷 label. This fallback is best-effort and optional: if pdftotext is not on PATH (nor at the usual Homebrew//usr/bin locations) the tool silently degrades to pdf.js-only. Install it via poppler (brew install poppler, apt install poppler-utils) to enable it.

update_books_metadata

Updates metadata for one or more books in a single transaction. Only the fields you provide are changed; omitted fields keep their current values.

get_book_tags

Gets the tags for a book.

set_book_tags

Replaces all tags for a book with the provided list. Pass an empty array to remove all tags.

Example workflow

List books with missing metadata, read the first few pages of each PDF,
infer the title and authors from the content and file path, then update
the metadata for all of them.

Claude will call list_books_needing_metadata, then read_pdf_pages for each book, and finally update_books_metadata to write everything in one transaction.

When the front matter is ambiguous, read_pdf_colophon recovers the ISBN, first-edition date, and publisher from the colophon at the end of the book — for Japanese books that is often the most reliable single source, and the extracted ISBN can seed a precise techbook-mcp lookup.

Combining with techbook-mcp

techbook-mcp is a companion MCP server that searches Japanese technical book metadata across multiple publishers (技術評論社, 達人出版会, 技術書典, etc.).

When both MCP servers are active in Claude Code, you can ask Claude to fill missing metadata by cross-referencing techbook-mcp search results:

list_books_needing_metadata でメタデータ不足の本を一覧して、
techbook-mcp で書名検索し、一致した書誌情報を update_books_metadata で書き込んで

Claude will:

1. Call list_books_needing_metadata to find books missing title or authors
2. For each identifiable book, call techbook-mcp:search_books with the title (inferred from filename or PDF content)
3. Confirm the match and call update_books_metadata to write the title, authors, publisher, and release date in a single transaction

Setup

Add both servers to your Claude Code project config (.mcp.json):

{
  "mcpServers": {
    "riida": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "riida-mcp"]
    },
    "techbook-mcp": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@zonuexe/techbook-mcp"]
    }
  }
}

Tips

• techbook-mcp covers Japanese technical publishers. For books not indexed there, use read_pdf_pages to extract metadata directly from the PDF content.
• BOOKSCAN-style filenames (書名 著者名 ページ数_ISBN.pdf) often contain enough information to fill title and authors without any search.
• Magazine split-files (e.g. Software Design 総集編) can be batch-updated from the directory path alone — no search needed.

Database path

The server automatically connects to the riida database:

License

MPL-2.0