Every other Anki MCP server depends on AnkiConnect, which was built for browser extensions and brings a port, an extra install step and an API that can change under you. Going straight at pylib trades UI integration for stability — a deliberate, stated tradeoff, and the right one if you mostly want cards created and corrected.
An MCP server for Anki that integrates with Anki's stable pylib API instead of the AnkiConnect add-on. Nothing to install inside Anki, no HTTP server occupying a port, no add-on to break — just Anki on the machine and an LLM client.
- Core deck operations — create, read, update and delete, with `create_note` for new cards and `list_collections` for what it found
- Automatic discovery of your Anki collections at their standard locations on macOS, Linux and Windows, with no paths to configure
- Sync — `sync_collection` against AnkiWeb by default or a self-hosted endpoint, with media included unless you set `sync_media: false`
- Credential handling — `save_sync_credentials`, `load_sync_credentials` and `delete_sync_credentials` keep the username and password in the OS credential store: Keychain on macOS, Credential Manager on Windows, Secret Service on Linux
- A default sync endpoint can be set in `config.json`; leave it null for AnkiWeb
Anki installed, Python 3.10+, and an MCP client. Run it with `uvx mousetail` — no add-ons and no dependencies beyond that. Close the Anki desktop application before syncing, or the sync fails. Two limits are by design: it will never integrate with the Anki UI, and it only works with an LLM tool on the same machine as the deck. Package version 0.1.5. MIT.
One command — uvx mousetail