Fluent

List evidence, provenance, source snapshots, event history, or evidence gaps for a subject or claim. For Meals recipe-to-grocery reasoning, pass the saved recipe ID or exact saved recipe title as subject to ground ingredients before proposing a grocery-list delta. Evidence can inform host reasoning, but it is not a final plan, style verdict, medical judgment, or checkout action.

fluent_list_evidence

List typed domain items such as Meals recipes, the living grocery list, inventory items, and Style closet items. Results are incomplete whenever nextCursor is returned: never conclude that an item is absent until relevant pages are exhausted or a confident match is found. Prefer targeted query values before broad inventory traversal. For saved Meals recipe discovery, pass item_type="recipe" plus query for the recipe title or ID, then call fluent_get_item with the returned ID before deriving grocery-list deltas. For Style photo ingestion, search discriminating brand or graphic text, garment description, and category/color evidence before creating; a matching item without Catalog media is an incomplete existing item to repair, not a new item. For owned-item outfit or shoe advice, list the relevant active Style category before recommending a saved item, then visually inspect shortlisted items through focused fluent_get_media_bundle calls.

fluent_list_items

Record explicit feedback for one saved Meals recipe and return read-after-write proof. This is recipe evidence, not a global preference, allergy, inventory, grocery, cart, or order update.

fluent_record_recipe_feedback

Refresh selected profile fields or typed natural-language wear feedback on an EXISTING saved Style closet item from explicit host/user evidence. This is the durable home for tags, descriptors, fit/wear learning, and recommendation corrections. Fluent rank-merges each field, preserves stronger existing evidence, never infers unworn from silence, and returns read-after-write proof.

fluent_refresh_style_item_profile

Render the current Fluent grocery-list MCP Apps surface. Use surface="meals_grocery_list". If the host cannot mount the returned ui:// resource, answer from the structured grocery-list fallback data in text.

fluent_render_surface

Save one explicit user-approved, host-authored Meals plan, derive its grocery plan from the referenced saved recipes, and return read-after-write proof for both the meal plan and living grocery list. Use only after the host model drafts the plan in conversation and the user approves saving it. This does not browse retailers, mutate carts, save recipes, or infer pantry/inventory truth.

fluent_save_meal_plan

Save one user-approved Meals recipe through the recipe validator and return read-after-write proof. A recipe without ingredients or instructions is saved as a draft and kept out of active meal planning until completed; grocery coverage depends on saved ingredients. Use only after the user explicitly approves saving the recipe.

fluent_save_recipe

Set one inspected source image or host-generated Catalog display image for a saved Style closet item and return read-after-write proof. In ChatGPT, pass an uploaded, selected, or generated image through image_file. Set catalog_ready=true only after reviewing the final primary Catalog presentation; for host-generated media, also provide source_photo_id for the exact retained non-generated source on this item. Fluent stores and hashes inline/hosted-file bytes and atomically binds the reviewed Catalog to its source. It never inspects images or browses product pages. A fit or alternate photo is ADDED by default (every saved photo and the cover are kept; repeating the same image and role is a no-op); only a primary photo or an explicit photo_action="replace" replaces its slot.

fluent_set_style_item_image

Render adapter for the Fluent Closet MCP Apps surface. Reads owned Style items and returns one collection-to-detail experience plus compact structured fallback data. When the user asks to review, compare, or resolve a possible duplicate between two saved closet records, call this tool with exactly those two saved IDs in filter.item_ids, filter.status="active", and presentation.mode="comparison"; omit focused_item_id. Do not claim that the comparison is open unless this tool was actually called and returned. After an outfit or shoe recommendation names one winning owned item, render the conversational payoff with filter.item_ids containing only its exact saved ID, presentation.mode="recommendation", focused_item_id set to the same ID, and one concise recommendation_reason; never substitute a category/subcategory/query filter or the whole closet. For an outfit of 3 to 5 exact owned items, render only those selected saved IDs with presentation.mode="comparison" and omit focused_item_id; the model must name the exact items and may offer at most one exact-owned replacement. Use presentation.mode="ingestion_review" only as the text-only/non-UI fallback when fluent_create_style_item could not attach its create-owned review app: follow fluent_create_style_item payload.reviewHandoff with filter.status="active", exact filter.item_ids, and the exact focused_item_id. Newly created items are active; pending_review is not a valid lifecycle or filter status. Exact filter.item_ids otherwise default to comparison and always provide an in-app route back to the full closet. Use mode="detail" with focused_item_id only when the user explicitly asks to inspect or manage the complete saved item. Category/type/brand/color/size/query filters are for deliberate collection browsing; status="archived" shows archived items. This surface manages saved closet state and hands exact stored item IDs to outfit creation; it does not invent garments or make purchase judgments.

fluent_render_style_closet_surface

Fetch a compact context packet for a domain and intent. Use this first for broad Meals planning, currentness checks, "what Fluent knows", and weeknight meal planning: call fluent_get_context with domain="meals" and intent="planning". Also use it before any closet-grounded "what should I wear?", "which shoes?", or owned-item outfit answer: call fluent_get_context with domain="style" and intent="closet" before naming items the user owns. The host model owns reasoning and final judgment; Fluent supplies durable context, typed items, evidence gaps, freshness, and suggested writeback boundaries.

fluent_get_context

Apply a typed sparse patch to an existing saved Meals recipe and return read-after-write proof. This cannot change recipe identity or create grocery/cart/order state.

fluent_update_recipe_patch

Apply an explicit, provenance-backed profile patch through the canonical shared or domain profile service. Use only when the user intends to change durable Fluent memory.

fluent_update_shared_profile_patch

Update one saved Style item with read-after-write proof. Use patch for catalog details, product_enrichment with an empty patch for a linked product and attributed material/care facts, or photo_library for photo arrangement. Tags and styling descriptors belong in fluent_refresh_style_item_profile. For user-approved closet management only; Fluent does not browse retailers or read email.

fluent_update_style_item_patch