[GUI Plugin] MCP Server Recherche – AI-powered research chat for your Calibre library

[Miguel Iglesias]

MCP Server Recherche turns your Calibre library into an AI-powered research assistant.
Instead of manually searching by title or tags, you can ask natural language questions and get focused book suggestions – with short summaries, sources, and direct links into your own library.

Ask things like:
– "Show me practical books on in-vehicle communication, especially CAN and FlexRay."
– "Which English books explain basic statistics for engineers?"
– "Find books that cover both battery management and vehicle electrical networks."

The plugin talks to a local MCP server that reads Calibre’s metadata database and uses AI to search, rank and summarize the most relevant books for your question.

FEATURES

• Natural-language search over your own Calibre library
  Type your question in plain English or German and let the AI figure out what to look for in your collection.
  
• Cross-language search with AI-generated keywords
  The plugin can automatically derive English keywords from German queries (and vice versa) to find books in a second language, even if your original query is monolingual.
  
• AI-generated summaries with clear source links
  For each matching book, the MCP server can fetch excerpts and generate a short, focused summary that explains why this book is relevant – with an indication of which passages were used.
  
• Mark relevant books directly in Calibre
  Matching books can be marked in Calibre (using the standard "marked" mechanism), so you can quickly come back to them later.
  
• Export results as JSON
  The list of suggested books – including titles, ISBNs and short descriptions – can be exported as JSON for further processing or custom scripts.
  
• Follow-up questions with full context
  You can ask follow-up questions like in a normal chat:
  – "Only show the introductory books."
  – "Which of these are in English?"
  – "Are there any titles that compare LIN and FlexRay?"
  The conversation keeps the context of your previous questions and results.
  
• Designed for real-world Calibre collections
  The plugin uses your existing Calibre metadata and works on top of

  metadata.db

  (read-only). It is meant for everyday use: open the window, ask a question, get useful books – no need to manually read everything first.

Requirements

• A Calibre library where the ISBN is stored in the "IDs" field for the books you want to use with the AI. This is currently a hard requirement so that the MCP server can reliably match books and fetch excerpts.
• An AI model accessible via API (tested primarily with OpenAI GPT-4 mini). Other models may work as well – suggestions are welcome.

Project & development

The MCP server and GUI plugin are open source and under active development.
GitHub repository: https://github.com/Miguel0888/mcp-server (replace with your actual repo URL)

Feedback, feature requests and pull requests are very welcome – or just fork the project, adapt it to your workflow, and enjoy finally finding the right books in your Calibre library.