The Scryfall MCP server defines 7 tools with complete input schemas and reasonable descriptions. However, several gaps limit its production readiness: (1) descriptions are inconsistent in depth and some lack actionable context; (2) no output schemas are documented, forcing LLMs to infer result structure from examples; (3) parameter descriptions are minimal and lack format/constraint guidance; (4) no error recovery guidance is provided; (5) missing tools for common follow-up queries (e.g., card price trends, set information). The server implements basic error handling but does not guide LLMs on recovery strategies. Tool naming is clear and verb-based (search_, get_), but some tools have overlapping functionality (get_prices_by_id vs get_prices_by_name) without explaining when to use each. Overall, this is a functional but baseline-quality implementation suitable for simple card lookups but lacking polish for complex agent workflows.
Retrieve a card by its Scryfall ID (a 36-char UUID). Returns the card data in JSON.
Retrieve a card by its exact English name, e.g. 'Black Lotus'. Returns the card data in JSON. If multiple cards share that exact name, Scryfall returns one (usually the most relevant printing).
Retrieve price information for a card by its Scryfall ID. Returns JSON with usd, usd_foil, eur, tix, etc.
Retrieve price information for a card by its exact name. Returns JSON with usd, usd_foil, eur, tix, etc.
Retrieve official rulings for a specified card by Scryfall ID or Oracle ID. Returns an array of rulings. Each ruling has a 'published_at' date and a 'comment' field.
Retrieve a random Magic card from Scryfall. Returns JSON data for that random card.
No output schemas documented. Tool descriptions mention 'Returns JSON' but do not document the actual fields, types, or structure agents should expect. This forces LLMs to infer schema from examples and risks downstream failures when chaining tools.
Parameter descriptions lack actionable constraints. 'A full text query' and 'The Scryfall UUID' tell agents WHAT fields exist but not HOW to use them. Missing guidance on valid query syntax (e.g. 't:goblin pow=2 o:haste'), UUID format validation, or what happens with invalid input.
Duplicate tool pairs (get_prices_by_id vs get_prices_by_name, get_card_by_id vs get_card_by_name) lack disambiguation guidance. Descriptions do not explain when to use each variant, forcing LLMs to reason about identity resolution or guess which is appropriate.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 55 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 51 | - | v1 |
Search for MTG cards by a text query, e.g. 'oracle text includes: draw cards'. Returns a list of matching cards (with basic fields: name, set, collector_number, ID). If no matches are found, returns an error message from Scryfall.
Error handling returns raw Scryfall API error JSON (e.g. 'Scryfall error: ... (code=..., status=...)') without recovery guidance. LLMs receive 'not_found' or 'invalid_expression' errors but have no suggested next action (e.g. 'Try search_cards() with a simpler query').
No pagination support documented or implemented. search_cards may return hundreds of results; without limit/offset parameters or result count, LLMs risk context overflow and cannot iterate over large result sets.
search_cards returns minimal fields (name, set, collector_number, ID) but does not specify the exact output structure or whether additional fields are included. Agents cannot confidently chain search_cards → get_card_by_id without inferring that 'ID' maps to the 'id' parameter.