calibre-mcp

A local MCP server (stdio) that lets an LLM host — Claude Desktop, Claude Code, or any MCP-compatible client — manage a Calibre ebook library conversationally: search, edit metadata, add, convert, deduplicate, remove, and email books, all with human-in-the-loop safety.

I built this to stop clicking around the Calibre GUI and manage my library from a chat instead. It's a personal project, and also a deliberate showcase of how I build backend/tooling systems: hybrid I/O design, an explicit failure taxonomy, human-approval gates on every destructive action, and a test suite that never touches real user data. See ARCHITECTURE.md for the full design writeup.

Why the hybrid design

• Reads (search, list, view, duplicate-finding) query metadata.db directly, read-only — fast, and structurally incapable of corrupting the library (the SQLite connection is opened mode=ro).

• Writes (edit, add, remove, convert, email) go through Calibre's own CLI tools (calibredb, ebook-convert, calibre-smtp) — never raw SQL — so Calibre stays authoritative over its own database.

• Every write is preceded by an automatic metadata.db backup (rolling, keeps the last 20).

• Removal is recoverable: files are copied to a managed trash folder and the book is sent to Calibre's recycle bin — never a permanent delete.

• Every mutating or outward-facing tool is two-step (plan → confirm): the first call returns a human-readable review plus a confirmation_token; nothing changes — and nothing is sent — until you re-call with that exact token.

Full rationale, module boundaries, and the decision log behind these choices live in ARCHITECTURE.md.

Related MCP server: calibre-mcp

Requirements

• Calibre installed, with calibredb and ebook-convert on your PATH (calibredb --version). calibre-smtp is also required if you want email_book.

• Python ≥ 3.12 and uv.

Install

git clone <this-repo> calibre-mcp
cd calibre-mcp
uv sync

Configure

The server manages one library, set via environment variable:

The server fails fast at startup with a clear error if CALIBRE_LIBRARY_PATH is unset or the directory has no metadata.db.

email_book additionally needs SMTP relay credentials (loaded lazily — the server boots fine without them, and only email_book fails if they're missing):

Claude Desktop / Claude Code

Add to your MCP config (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "calibre": {
      "command": "uv",
      "args": ["--directory", "/absolute/path/to/calibre-mcp", "run", "calibre-mcp"],
      "env": {
        "CALIBRE_LIBRARY_PATH": "/absolute/path/to/your/Calibre Library"
      }
    }
  }
}

Run manually

CALIBRE_LIBRARY_PATH="/path/to/Calibre Library" uv run calibre-mcp
# or equivalently:
CALIBRE_LIBRARY_PATH="/path/to/Calibre Library" uv run python -m calibre_mcp

The server communicates over stdio (JSON-RPC); it prints nothing to stdout except MCP framing — all logs go to stderr, on purpose (see ARCHITECTURE.md).

Tools

Every tool's full contract (edge cases, error conditions, exact field whitelist) is documented in its docstring in server.py — those docstrings are what the LLM host sees, so they double as the API reference.

Development

uv run ruff check src tests   # lint
uv run pytest                 # full suite (unit + integration + e2e)
uv run pytest -m unit         # fast unit tests only

137 tests across three tiers (unit, integration, e2e); write tests never touch a real library — see ARCHITECTURE.md.

Project layout

src/calibre_mcp/
├── server.py               # FastMCP tool surface — the only stdio/MCP-aware module
├── library.py               # CalibreLibrary facade — orchestrates every tool's business logic
├── sqlite_reader.py         # Read-only metadata.db access (the only sqlite3 call site)
├── calibredb_runner.py      # calibredb subprocess wrapper (search/edit/add/remove/add_format)
├── ebook_convert_runner.py  # ebook-convert subprocess wrapper
├── calibre_smtp_runner.py   # calibre-smtp subprocess wrapper
├── backup.py                 # metadata.db snapshots + recoverable trash
├── confirmation.py           # plan→confirm token derivation/verification
├── config.py                  # env-driven startup config, fail-fast validation
└── errors.py                  # the failure taxonomy every layer maps to

Roadmap

Shipped: full read/curate/distribute loop (search, list, view, edit, add, remove, convert, dedupe, email). What's next — library self-knowledge, bulk operations, cover/metadata enrichment, device sync — is tracked in .specs/ROADMAP.md, including the reasoning for sequencing and what's explicitly out of scope.

Contributing

See CONTRIBUTING.md for the dev workflow, the invariants a PR must preserve, and how the spec-driven process behind this repo works.

License

MIT © Gustavo F Sousa.