epub-mcp-server

Description

epub-mcp-server is a Model Context Protocol (MCP) server that lets AI agents create, read, and edit EPUB novels. It exposes tools for reading and writing an EPUB's metadata, chapters, navigation, cover, spine, manifest, guide, and other resources — effectively giving an LLM the ability to understand and manipulate EPUB structure directly.

Written in TypeScript, it runs on Bun or Deno using the stdio transport protocol, making it suitable for integration with any MCP-compatible client such as Claude Desktop.

Related MCP server: tomato-writer-mcp

Table of Contents

• Installation

  • Prerequisites: Node.js

  • Option A — Direct from GitHub (Bun or Deno)

  • Option B — Global NPM Install

• Usage

• Features

Installation

Prerequisites: Node.js

Node.js is required for Option B (Global NPM Install). If you plan to use that option, install it first using one of these methods:

• Windows / macOS: Download the LTS installer from nodejs.org and run it.

• Linux (apt): curl -fsSL https://deb.nodesource.com/setup_lts.x | sudo -E bash - && sudo apt-get install -y nodejs

• Homebrew (macOS / Linux): brew install node

Verify the installation by running node --version and npm --version in your terminal.

Option A — Direct from GitHub (Bun or Deno)

If you have Bun, Deno, or both installed, you can run the server directly from raw.githubusercontent.com without installing it locally. No compilation or OS code-signing warnings needed.

⚠️ Rate-limit notice: This path fetches the server from raw.githubusercontent.com every time your MCP client starts. GitHub enforces an anonymous usage policy that limits unauthenticated requests to 60 per hour (across all of github.com and its subdomains). If you exceed this limit, requests will be rejected with a 403 Forbidden error until the window resets. Frequent restarts can trigger this — for heavy use, see Option B which caches everything locally.

Option B — Global NPM Install

If you prefer a traditional Node.js/NPM setup, install the package globally from the npm registry. This downloads all dependencies locally so the server boots instantly and works 100% offline.

npm install -g epub-mcp-server

To update later: npm update -g epub-mcp-server

Because this installs a pre-built package from the registry (rather than cloning and building the repo locally), it avoids the Windows npm/node-tar ENOENT race that git-based (github:user/repo) installs are prone to.

Usage

Add a configuration block to your MCP client's config file (e.g., claude_desktop_config.json or .mcp.json). Select the entries that apply to you:

Using Bun only:

{
  "mcpServers": {
    "epub-mcp-server-bun": {
      "command": "bun",
      "args": ["run", "https://raw.githubusercontent.com/dmongrel/epub-mcp-server/master/src/index.ts"]
    }
  }
}

Using Deno only:

{
  "mcpServers": {
    "epub-mcp-server-deno": {
      "command": "deno",
      "args": ["run", "--allow-env", "--allow-read", "--allow-write", "https://raw.githubusercontent.com/dmongrel/epub-mcp-server/master/src/index.ts"]
    }
  }
}

Using both Bun and Deno:

{
  "mcpServers": {
    "epub-mcp-server-bun": {
      "command": "bun",
      "args": ["run", "https://raw.githubusercontent.com/dmongrel/epub-mcp-server/master/src/index.ts"]
    },
    "epub-mcp-server-deno": {
      "command": "deno",
      "args": ["run", "--allow-env", "--allow-read", "--allow-write", "https://raw.githubusercontent.com/dmongrel/epub-mcp-server/master/src/index.ts"]
    }
  }
}

Using global NPM install:

{
  "mcpServers": {
    "epub-mcp-server": {
      "command": "epub-mcp-server",
      "args": []
    }
  }
}

Features

28 MCP tools organized across 10 categories:

EPUB Lifecycle

• new_epub — Create a blank EPUB on disk with title, author, and cache it in memory.

• read_epub — Parse an existing .epub from disk; returns metadata, reading order, and TOC (cached via LRU).

• save_epub — Write all cached edits back to disk atomically.

• close_epub — Free a cached EPUB's memory slot (discards unsaved edits unless already saved).

• reload_epub — Discard cache and re-parse from disk, restoring a clean state.

• get_cache_status — List all EPUBs currently held in memory with their dirty flags.

• get_epubs_list — Scan a directory for .epub files (with optional recursion).

Content & Chapters

• get_chapter — Read one content document by its internal id, returning both plain text and raw XHTML markup.

• edit_chapter — Create, edit, or remove chapters. Supports markdown with auto-chapter splitting and full XHTML input.

• convert_manuscript — Convert an entire .txt, .md, or .html manuscript into EPUB chapters in one call, splitting on chapter markers.

• find_text — Search chapter prose (excluding front/back cover pages) for a plaintext substring or regex pattern, reporting the chapter and line number of every match. Optionally limit the search to specific 1-based chapter numbers.

• validate_epub — Check an EPUB for misalignment between its table of contents, spine, manifest, and chapter text (e.g. a toc entry labelled "Chapter 5" pointing at chapter 7), plus dangling references, duplicate ids, malformed XHTML, and missing required metadata. Read-only; every finding names the tool call that fixes it. Optionally limit the run to specific checks.

Metadata (Dublin Core)

• get_metadata — Read all metadata fields: identifiers, titles, languages, creators, contributors, publishers, dates, subjects, description, rights, and custom metas.

• edit_metadata — Create, edit, or remove individual metadata entries with rich attribute support (scheme, role, lang, refines, etc.).

Covers

• get_cover — Read the front cover image; returns base64 data or writes directly to disk. Supports EPUB 3 cover-image and legacy EPUB 2 meta pointers.

• edit_cover — Create (adds manifest item, XHTML wrapper, spine entry, landmark, guide reference), replace, or remove the front cover.

• edit_back_cover — Create, edit, or remove a back cover image with spine/landmark/guide wiring.

Spine & Reading Order

• get_spine — Read reading order including page progression direction (ltr/rtl), linear flags, and properties.

• edit_spine — Add, reorder, or remove spine entries; adjust linear flag and properties.

Manifest

• get_manifest — List every manifest item with id, media type, properties (cover-image, nav, scripted), fallbacks, and media-overlay references.

• edit_manifest — Modify existing items' media type, properties, or fallback/media-overlay pointers.

Guide (Legacy EPUB 2)

• get_guide — Read legacy guide landmarks (type, title, href).

• edit_guide — Create, edit, or remove individual guide references by type.

Navigation & TOC

• get_navigation — Read navigation document: table of contents, landmarks, and page-list as a typed tree structure. Includes NCX presence flag.

• edit_navigation — Create nested entries, edit labels/hrefs/types, or remove entries with their children. Auto-syncs legacy NCX for TOC changes.

Resources (Stylesheets, Fonts, Media)

• get_resource — Read any non-content manifest resource as text or base64; write directly to disk via sourcePath.

• edit_resource — Create (adds to manifest, not spine), replace, or remove resources. Media type auto-guessed from extension.

Utility

• get_context — List all available tools with descriptions and extended usage guidance.