设计与多媒体

Bring Photo MCP

试用

Turn product photos into verified Bring! list items via MCP.

它能做什么

Use this skill for Bring shopping-list operations that need catalog-based classification, product-photo handling, and verified list writes.

技能文档

Bring Photo MCP

When to Use

Use this skill for Bring shopping-list operations that need catalog-based classification, product-photo handling, and verified list writes.

Boundary

  • Perform deterministic list operations exclusively through the bring_* MCP tools provided by this server.
  • Do not construct, replay, inspect, or send raw Bring HTTP requests. Do not use browser automation, shell commands, or a second Bring client for operations covered by these tools.
  • Never expose credentials, list IDs, item-detail UUIDs, image paths, or photo bytes in conversation output.
  • Treat every tool response as authoritative for the operation it reports. Do not infer success from intent.

Photo-led registration workflow

  1. Analyse the supplied product photo locally in the conversation. Extract only purchase-relevant facts: visible product name, brand, variant, amount, and ambiguity. If the image is unreadable or ambiguous, ask for clarification; do not write.
  2. Search the configured locale catalog with bring_search_catalog({ query, locale?, limit? }) using the visible main name first, then specific-to-general terms if needed.
  3. Decide catalog versus free text:
    • For an exact catalog product, use the catalog main name and its returned icon/category pair.
    • For a product that is not an exact catalog item, retain an identifying free-text item_name, but select an explicitly returned catalog icon/category pair.
    • Never assume Eigene Artikel is a valid API category.
  4. If the catalog search does not make the classification clear, call bring_suggest_classification({ product_name, hints?, locale? }). A response with insufficient confidence is a no-write result: present candidates or ask for clarification.
  5. Compose specification at no more than 30 Unicode code points. Do not repeat the main/catalog name. Prefer brand + distinguishing variant + amount, keeping the product identifiable. If a required fact is not visible, do not invent it.
  6. Enforce the hard icon/category invariant: every bring_add_item call must contain icon_item_id and category_id from the same valid catalog entry. No create, copy, or move is successful without verified persisted icon and category.
  7. Call bring_add_item({ lists, item_name, specification, icon_item_id, category_id, image_path?, duplicate_policy }). Add image_path only when the user supplied a permitted local photo path.
  8. Report the returned persisted fields and verification evidence per list, without repeating sensitive identifiers or local paths.

Duplicate policy and idempotency

  • Use duplicate_policy=fail by default. An ambiguous existing item is an explicit no-write result, not a reason to overwrite.
  • Use duplicate_policy=update_exact only when the intended existing item has exactly the same item identity and specification.
  • Repeated identical requests should converge on one verified item per target list. Never silently overwrite another active item with a generic shared name.

Photo operations

  • Use bring_attach_photo, bring_replace_photo, or bring_remove_photo only for a confirmed target item.
  • Trust success only when the tool returns its read-back verification (imageUrl present after add/replace; absent after removal).
  • For images supplied as a conversation attachment rather than a local path, obtain an approved local path through the host workflow before calling a photo tool; do not fabricate a path.

Copy and move

  • For copies, call bring_copy_item({ source_list, target_lists, item_name, include_photo }). Metadata must remain catalog-valid; the source remains unchanged.
  • For moves, call bring_move_item({ source_list, target_list, item_name, include_photo }). The target must be completely verified before the source is removed. If target creation or verification fails, report failure and leave the source untouched.
  • For every multi-target add or copy, enumerate the per-list results. If any result fails and any succeeds, report status: "partial"; never call it total success. For all failures, report status: "error" with stable error codes.

Other deterministic operations

  • Discover allowed lists with bring_list_lists and inspect them with bring_list_items.
  • Verify an existing item with bring_get_item before a sensitive follow-up operation.
  • Use bring_check_item, bring_uncheck_item, and bring_remove_item only on the confirmed requested list and item.

Safe response shape

For writes, state: requested operation, overall status, then one concise result per list (success/partial/error, stable error code when present, and non-sensitive verification evidence). State clearly when no write was performed due to ambiguity, insufficient classification confidence, or duplicate policy.

相关技能

把自然语言描述转为结构化 JSON,并由 mcp-diagram-generator MCP 服务生成 Draw.io、Mermaid 或 Excalidraw 图表文件。

作者 nssa.io1.0k 次安装47 星标

诊断生产力系统反复失效的根因,给出最小干预——容量测算、瓶颈定位、可靠的本地记录。

作者 Iván854 次安装69 星标

以 AI 机器人身份加入视频会议,提供语音、虚拟形象与屏幕共享四种模式。

作者 johnpatternai21 次安装8 星标

从 AdMapix API 拉取广告创意、应用、榜单和收入预估等数据,原样返回结构化 JSON。

作者 fly0pants4.3k 次安装296 星标

在本地磁盘以分类纯 Markdown 文件保存需要长期留存的事实,与智能体内置记忆并存。

作者 Iván555 次安装18 星标