Developers
fav.ing MCP connector
fav.ing is a social list-making app where people share recommendations from people they trust: movies, books, restaurants, music, products, local pros and more. The connector lets an AI agent answer “what should I watch, eat, read or buy?” with picks from real people the user follows, and save things to the user’s lists, bookmarks and current favorites.
Endpoint
MCP over Streamable HTTP: https://fav.ing/api/mcp
Authentication
- OAuth 2.1 authorization code flow with PKCE (S256). Every request carries the user’s bearer token.
- Dynamic client registration is supported. Discovery starts at
https://fav.ing/.well-known/oauth-protected-resource, which names the authorization server. - Grants:
authorization_codeandrefresh_token. - A free fav.ing account is all a user needs. They sign in and approve access during the OAuth flow.
Tools
Every tool carries MCP annotations: reads are marked readOnlyHint: true, writes are not. set_current_favorite and unfollow_user are marked destructiveHint: true (one replaces the current pick, the other removes a follow), and writes that are safe to repeat are marked idempotentHint: true. That is the connector’s read/write split, so an agent can ask for read access alone.
| Tool | Access | What it does |
|---|---|---|
| get_recommendations | Read | Picks from the people the user follows, or from one named member (`from`), filtered by what they want, a category and a place (`near` coordinates, or place words). Words must actually match (genres count: 'sci-fi' finds Science Fiction); members rank above fav.ing's editorial accounts. Falls back to the wider fav.ing community, labelled as such, and returns nothing rather than unrelated picks. |
| get_current_favorites | Read | Each person's one current favorite per category (movie, album, restaurant…), for the people the user follows or for the user. |
| get_trending_lists | Read | Public lists with the most likes, comments and re-lists this week or month, optionally around a place (`near`). If nothing was engaged with, it widens to 30 days, then the newest lists, and says which window answered. |
| get_list | Read | Every item on one list, with the author's notes, from a fav.ing list URL or id, or an owner's @username plus the list's name. |
| suggest_people_to_follow | Read | People worth following, optionally for a category or a place: creators of matching public lists, or fav.ing's own taste-based picks. Never people the user already follows or has blocked. |
| search_lists | Read | Find the user's own lists by title. |
| create_list | Write | Create and publish a new list on the user's profile, with its items (and optionally a place) in one call. |
| add_item | Write | Add one item to one of the user's lists. |
| save_bookmark | Write | Save a thing, a link, a place or a whole fav.ing list (by URL, or @owner plus its name) to the user's bookmarks. Bookmarks are private. |
| suggest_item | Write | Suggest an item for another member's list that is open to suggestions. The owner is notified; an earlier identical suggestion is upvoted instead. |
| import_list_from_url | Write | Turn an article or page that lists things into a new list on the user's profile, using their fav.ing import allowance. |
| set_current_favorite | Write | Set the user's current favorite in a category. It replaces the current one, which moves to their history. |
| mark_done | Write | Mark something the user watched, read, ate at or tried as done, or undo it. Safe to repeat. |
| follow_user | Write | Follow a member by @username. They're notified, as in the app. Safe to repeat. |
| unfollow_user | Write | Stop following a member. Safe to repeat. |
Places
get_recommendations, get_trending_lists and suggest_people_to_follow take an optional near: { lat, lng, radius_km } (radius defaults to 40 km). The agent supplies the coordinates of the place the user names; fav.ing matches the places on each list, suburbs included. Plain place words (location) still work as a fallback.
Links and affiliate disclosure
Product links may be affiliate links; creators may earn a commission. An earning link is a https://fav.ing/out?… redirect that credits the person whose list or current favorite it came from, then sends the shopper straight to the store. Every other link is the item’s own URL with tracking parameters removed.
Privacy
- Every call runs as the signed-in user under the database’s row-level security.
- Other people’s content is limited to public, published lists and public current favorites. Private, followers-only, invite-only and link-only lists are never returned to anyone but their owner.
- Writes only ever change the signed-in user’s own lists, bookmarks, current favorites, done items and follows. The one exception is
suggest_item, which adds a suggestion to another member’s list only when that list is open to suggestions. - We don’t sell personal data. Images an agent sends are copied into fav.ing’s own storage.
- Privacy Policy · Terms of Service
Limits
- Per user: 120 reads and 30 writes a minute. Over the limit, a tool returns an error asking the agent to wait.
- Read tools return at most 20 results;
get_listreturns up to 100 items. - Available worldwide; content is mostly English.
Support
Email support@fav.ing.