Calibre MCP Server

This MCP server bridges the gap between AI agents and your Calibre ebook libraries. It enables agents to interact with your collection as a dynamic knowledge base, allowing them to search, manage, and read your digital libraries. Unlike other MCP servers it can also allow an AI agent to update library metadata and contents if a library's permissions are set to allow it.

Key Features

• Advanced Search: Query libraries using book metadata or perform full-text content searches.

• Metadata Management: View and update metadata (titles, authors, tags, ratings, etc.) for any book.

• Library Maintenance: Add new titles to your collection or remove existing ones.

• Format Conversion: Leverage Calibre’s powerful conversion engine to switch between ebook formats (e.g., PDF to EPUB) on the fly.

• Direct Reading: Search and read the text content of a book directly into the agent's context window for analysis, summarization, or Q&A.

• Granular Permissions: Define strict access controls per library, including read-only modes and field-level write restrictions.

Prerequisites

• Calibre: Must be installed on the host system. This server utilizes calibre-debug to execute worker processes.

• Concurrency Note: Calibre does not support concurrent access to a single library. Do not point an agent to a library currently being used by the Calibre desktop application or other calibre processes to avoid database corruption. Setting a worker_timeout can reduce the risk of this happening accidentally, but don't rely on that to protect your libraries. Be aware of what you're doing.

Configuration

The server is configured via a JSON file. Since JSON does not support comments, use the structure below as a template.

Example Configuration (config.json)

JSON

{
    "libraries": {
        "default": {
            "path": "d:/ebooks/main_library",
            "description": "The primary research library containing technical manuals.",
            "default": true,
            "permissions": {
                "read": ["title", "authors", "tags", "rating", "comments"],
                "write": ["tags", "rating", "comments"],
                "delete": false,
                "convert": true
            },
            "import": {
                "allowed_paths": ["d:/downloads/ebook_imports"],
                "allow_delete_source": false
            },
            "export": {
                "allowed_paths": ["d:/ebook_exports"s],
                "allow_overwrite_destination": false
            },
            "worker_timeout": 300
        }
    },
    "port": 8000,
    "enable_worker_logging": false,
    "expose_resources_via_tools": true,
    "log_level": "warning"
}

Configuration Schema Reference

Each library is identified by a unique text key and can have its own separate configuration values:

There are also several top-level configuration settings that apply to the MCP server as a whole:

Example MCP Configuration

You can run this MCP server using uvx, which will seamlessly download and run the latest version from GitHub without needing a manual installation.

First, install uv if you haven't already.

Then, point to your configuration file with the CALIBREMCP_CONFIGPATH environment variable in your MCP clients configuration. For example:

JSON

{
    "mcpServers": {
        "calibre": {
            "command": "uvx",
            "args": [
                "--from",
                "github.com/FaceDeer/calibre_full_mcp_server",
                "calibre-full-mcp"
            ],
            "env": {
                "CALIBREMCP_CONFIGPATH": "<full path to your config.json>"
            }
        }
    }
}

(Note: If you have cloned the repository locally and wish to run from your local copy, you can replace "github.com/FaceDeer/calibre_full_mcp_server" with the <full path to your local repository directory>.)

This should allow you to easily switch library configurations as needed for different agents by selecting which configuration to point CALIBREMCP_CONFIGPATH at.

Pro-Tips for Better Performance

1. Optimize Field Access

Calibre libraries often contain internal metadata that can clutter an agent's context window. It is highly recommended to use a specific list for read permissions. Only expose fields the agent actually needs (e.g., title, author, tags, comments).

2. Custom Fields & Series

• Custom Fields: If you use custom columns in Calibre, you must include the # prefix (e.g., #my_custom_field).

• Series: Every series field has a corresponding series_index. Ensure both are included in your permissions list if you want the agent to see or manage book order within a series.

• Descriptions: The server passes custom field "descriptions" to the agent. Use these in Calibre to give the agent hints on how to use specific custom columns. For example if you create a custom field for a book's "age rating"" you could use the description to explain what the values of that field represent.

3. Resource Exposure

If your agent does not yet support the @mcp.resources standard, set expose_resources_via_tools to true. This will expose dedicated tools that allow the agent to fetch files via standard tool calls.

Docker Support

If you are building a Docker image for this server, it is recommended to pre-download the necessary NLTK models during the build phase to avoid runtime downloads and slow startup.

You can use the provided setup script for this purpose:

# Inside your Dockerfile
RUN calibre-full-mcp-setup

This will download the punkt and punkt_tab models into the image.

Architecture

For a deep dive into how this server manages worker processes and interacts with the Calibre database, please refer to the Architecture Documentation.