Unofficial MyFigureCollection.net API — Python library and local MCP server
Strong naming conventions (all verb_noun pattern), comprehensive descriptions (avg 180 chars, well within 10-1024 range), and complete input schemas with types and descriptions across all 12 tools. Pagination support present (page/limit params). However, output schemas are not explicitly documented in the code, only inferred from Pydantic model_dump() calls. Error handling converts MFCError to ValueError but lacks recovery guidance. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being read-only. Parameter descriptions are solid but could include format/constraint details (e.g., 'page must be ≥1').
Get an MFC club: description, member and comment counts, discussion threads, recent comments, staff, and a sample of members.
Get the full, paginated member roster for a club.
Get one page (50 items) of a user's collection. `status` is one of: owned, ordered, wished, favorites. Check `pagination.has_next_page` and call again with page+1 for more.
Get a single MyFigureCollection item by id. Returns the full item page: name, category, picture, origins, characters, companies, artists, releases (date/edition/price/barcode), scale, height, rating, and how many users own/ordered/wished it.
Get an item list by id: its owner, description, tags, and items.
Get the partner shops that sell an item, with prices where MFC has them. Returns every MFC partner shop with an availability of `available`, `not_available` or `maybe_available`, plus `price` and `currency` for the ones in stock. The `url` is MFC's affiliate redirect, not the shop's own product page. Important: real availability and prices only come back for items MFC has a barcode for. For an item with no barcode every shop reports `maybe_available` with no price — that means "not checked", not "in stock".
Output schemas not documented. Code uses Pydantic model_dump(mode='json') but no schema definitions visible in tool registration. LLMs cannot predict response structure for planning downstream calls.
No tool annotations despite all 12 tools being read-only. Missing readOnlyHint would help agents understand these are safe to call without side effects.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | <=2025-11-25 | v2 |
Get a user's public profile: join date, hits, rank, about fields, and per-category collection counts.
Get a shop: homepage, location, shipping, rating, review and item counts.
Get the item lists a user has published, with item counts and ids.
Find an MFC item by its JAN/UPC barcode. This is the join key for matching an item across stores. A unique barcode returns the full item under `item` with `matched: true`; no match returns `matched: false` with an empty `items` list. Note that MFC does not have a barcode for every item — plenty of items, especially Western releases, have none recorded, and cannot be found this way.
Search MFC items by title. 50 results per page. `category_id` optionally narrows by MFC category (1 = Prepainted, 2 = Action/Dolls, 3 = Trading, 10 = Model Kits, 21 = Books, ...).
Browse MFC's shop directory. 10 shops per page. `keywords` matches location as well as name, so "Japan" works. `average_score` filters to shops rated at least that (1-5). `partners_only` restricts to MFC partner shops — the same set that appears in get_partner_listings.
Error handling converts MFCError to ValueError but provides no recovery guidance. 'User not found' should suggest 'Try search_shops() or search_items() first.' Bare errors leave agents stuck.
Parameter descriptions lack format/constraint details. E.g., 'page' param has no min/max bounds stated. 'category_id' lists examples (1, 2, 3, 10, 21) but should be an enum or reference a discovery tool.
get_collection validates 'status' enum at runtime but parameter schema shows type 'string' with no enum constraint. LLMs cannot see valid options without calling the tool and failing.