AnkiMCP Server

The AnkiMCP Server connects Anki, the widely used flashcard software with spaced repetition, to AI assistants such as Claude Code or Cursor. Instead of typing and formatting cards by hand, you describe what you want to learn and the assistant creates matching cards in your real collection. The project is developed as a community addon and, by its own account, is not affiliated with Ankitects, the maker of Anki.

What is the AnkiMCP Server?

It is an addon for the Anki desktop application that starts a local MCP server. As soon as Anki is running, an HTTP endpoint is available at http://127.0.0.1:3141/ that MCP clients can connect to. The assistant can then read and change your decks, notes, cards, tags, and note types. Because the server runs inside Anki, it works with the same collection you use for studying every day, not with a copy.

Features at a glance

According to the README, the addon provides more than fifty tools. These include creating and editing notes, managing decks, tags, and note types, and searching the collection. There are tools for study sessions in which the assistant presents cards and rates answers, and support for optimising the parameters of the FSRS scheduler. The connection to the interface is also handy, for example to open a deck for review. The project site additionally mentions ready-made prompts, pronunciation audio for language learners, and a knowledge map called Anki Studio, which belongs to the paid offerings.

Setup and connection

The prerequisite is Anki version 25.07 or newer. You install the addon through AnkiWeb (code 124672614) or as an .ankiaddon file from the GitHub releases. For Claude Code and Cursor, the direct HTTP transport with the local address is enough; for Claude Desktop, the README uses the mcp-remote bridge via npx. If you want to connect ChatGPT or Claude.ai in the browser, you can use a tunnel, which the project offers as a relay or via ngrok. You then have to configure the allowed hosts and origins explicitly.

Security and privacy

By default the server checks Host and Origin headers against loopback addresses and thereby protects against DNS rebinding. For remote access, a bearer token can be set up. According to the provider, diagnostic logs redact secrets. Keep in mind nevertheless that the contents of your cards enter the model's context and therefore reach the respective model provider.

Costs and licence

The addon is open source under AGPL-3.0-or-later and free to use. According to its own site, the project additionally offers paid plans, for example with hosted Anki; current terms are on the official pricing page of ankimcp.ai.

Who is the AnkiMCP Server for?

It suits anyone who already uses Anki regularly, for example for languages, medicine, law, or programming, and who finds card creation the most tedious hurdle. An assistant can derive cleanly worded cards from notes, scripts, or documentation and act as a tutor while you review. If you do not use Anki, you gain nothing. And if you care about the highest card quality, check generated cards briefly before studying, because models can summarise facts incorrectly.

Frequently asked questions

Does the server run without Anki open? No, it is part of Anki and only active while the application is running.

Do I need a tunnel? Only for clients that cannot reach your machine locally, such as Claude.ai in the browser.

Is this an official Anki product? No, it is an independent community project.

Anki Desktop version 25.07 or newer with the AnkiMCP addon installed, plus an MCP client with HTTP support (Claude Code, Cursor) or the mcp-remote bridge for Claude Desktop.

With the AnkiMCP Server, an AI assistant works directly in your Anki collection. This guide leads from installation to the first request.

Step 1: Check the prerequisites

You need Anki Desktop version 25.07 or newer and an MCP client with HTTP support, such as Claude Code or Cursor. According to the README, older Anki versions with Python 3.9 are not supported.

Step 2: Install the addon

In Anki, open Tools → Add-ons → Get Add-ons and enter the code 124672614. Alternatively, download the .ankiaddon file from the GitHub releases. Restart Anki afterwards.

Step 3: Connect the client

As long as Anki is running, the server is reachable at http://127.0.0.1:3141/. Enter this address in Claude Code or Cursor as an HTTP MCP server. For Claude Desktop, the project uses the npx mcp-remote bridge.

npx mcp-remote

Step 4: Test the connection

Start with a read-only request, such as "Which decks are in my collection?". Then have two or three test cards created in a separate deck and check them in Anki.

Use remote access deliberately

For Claude.ai or ChatGPT in the browser, the project offers a tunnel. Configure allowed hosts and origins for it and set an API key so that third parties cannot access your collection.

Set limits

The assistant can change and delete cards. Back up your collection beforehand with Anki's backup function and review generated cards before you study them.

No sign-in needed locally; for remote access through a tunnel, an optional API key (bearer token) can be configured.

Full read and write access to decks, notes, cards, tags, and note types of the opened Anki collection, plus control of the study interface.

The server runs locally in Anki. Card texts the assistant retrieves enter the AI client context and therefore reach the model provider; when using a tunnel, traffic additionally passes through the relay or ngrok service.

Accidental deletion or overwriting of cards, an overly open tunnel without a token, misconfigured allowed hosts, and incorrect card content generated by the model.