Headless AnkiConnect-compatible REST API server with AnkiWeb sync and MCP support for AI assistants
Strong foundation with 35 well-named tools, comprehensive schemas, and excellent tool annotations. All tools have descriptions (avg 95 chars) and input schemas with typed parameters. Server instructions are detailed and actionable. However, parameter descriptions are often minimal (1-2 sentences), output schemas are not documented in code, and error handling guidance is absent. No pagination support despite tools returning lists (deck_names, model_names, tags). Missing idempotency hints for non-destructive writes. Tool composition is sound but some tools could batch operations.
Add a new note to the collection.
Add tags to notes.
Check if cards are due for review.
Check if cards are suspended.
Convert card IDs to note IDs.
Move cards to a different deck.
Create a new deck with the given name.
Delete one or more decks. Set cardsToo to True to also delete cards.
Output schemas not documented in code. Tools return structured data (lists, dicts, SyncResult) but no docstrings describe return types, field names, or structure. LLMs cannot plan downstream calls without knowing what fields to expect.
Parameter descriptions are minimal (1-2 words). E.g., 'query' for find_notes lacks guidance on syntax (is it free-text? Anki query language?). 'complete' for get_card_intervals lacks explanation of what 'complete' means. Descriptions should be 50-150 chars and explain WHAT, WHEN, and FORMAT.
No pagination support for list-returning tools. get_deck_names, get_model_names, get_all_tags, find_notes, find_cards return unbounded lists. Large collections could return thousands of items, exhausting context. Add limit/offset or cursor parameters and document max result size.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2026-07-28+ | v2 |
Delete a media file.
Delete notes by their IDs.
Export a deck to an .apkg file.
Find cards matching the given search query.
Find notes matching the given search query.
Get all tags in the collection.
Get the AnkiConnect API version.
Get intervals for cards.
Get detailed information about specific cards.
Get deck configuration.
Get all deck names in the collection.
Get all deck names with their IDs.
Get the path to the media directory.
Get all field names for a specific model.
Get all note model names.
Get CSS styling for a model.
Get card templates for a model.
Get detailed information about specific notes.
Get sync status from AnkiWeb.
Import an .apkg file.
Remove tags from notes.
Retrieve a media file.
Store a media file.
Suspend one or more cards.
Sync collection with AnkiWeb. Returns a structured SyncResult describing the collection outcome (no_changes / merged / downloaded), the download reason if any, the media sync counters, and any server message. Full uploads are never performed: if AnkiWeb is empty and requests a full upload, the call raises instead of replacing remote data.
Sync only media files with AnkiWeb (not the collection).
Unsuspend one or more cards.
No error handling guidance in tool descriptions. Tools like delete_decks, delete_notes, delete_media_file are destructive but lack recovery hints. Descriptions should state: 'If deck not found, call get_deck_names() to verify.' Error responses should include actionable next steps.
Idempotency not documented for non-destructive writes. create_deck, add_note, add_tags are marked IDEMPOTENT_WRITE or ADDITIVE_WRITE but descriptions don't state: 'Safe to retry. Duplicate calls with same input produce same result.' This guidance helps agents decide whether to retry on transient failures.