搜蚁智选 (AIMALL) 官方 MCP Server — 零依赖,纯标准库实现。让任何支持 MCP 的 Agent(Claude Desktop / Cursor / Dify / Coze / LangChain)一键接入搜蚁智选的大模型与工具市场,按量付费。
Server has 5 tools with complete input schemas and descriptions. Naming follows verb_noun pattern (list_, ai_, bootstrap_, get_). Descriptions are present but vary in quality (10-200 chars). All parameters have type definitions and descriptions. However, output schemas are not documented, error handling lacks recovery guidance, and no tool annotations (readOnlyHint/destructiveHint) are present. Parameters accept free-form strings where enums would be safer (e.g., api_type in list_assets has enum but ai_request's model param does not). No pagination support despite list_assets potentially returning many items.
在不扣费的情况下预估一次 AI 调用的费用(USD)。调用前用它做预算闸门。
调用搜蚁智选的某个 AI 资产(大模型对话或工具),按量计费并返回结果、订单号与余额。
(可选)自助注册账号并创建 Agent,返回可用的 API Key。会创建真实账号,仅首次接入时使用。
返回搜蚁智选的接入指引与机器可读文档链接。
列出搜蚁智选市场中的可用 AI 资产(大模型 / 工具),含 asset_id、类型与单价。先用它找到要调用的 asset_id。
No output schemas documented. LLMs cannot predict response structure or plan downstream tool calls. ai_request returns unstructured text with optional metadata fields; ai_estimate returns JSON but structure is not declared.
Error responses lack recovery guidance. Errors return '[error] <status>: <raw_json>' with no actionable next steps. E.g., 'API_KEY not set' should suggest 'Set AIMALL_API_KEY env var and restart' or 'Call bootstrap_account to create one'.
No tool annotations. ai_request is destructive (charges money) but lacks destructiveHint. list_assets and ai_estimate are read-only but lack readOnlyHint. Agents cannot distinguish safe from risky calls.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2026-07-28+ | v2 |
ai_request's 'model' parameter accepts free-form strings with no enum or validation. LLMs may hallucinate invalid model names. Should either enumerate valid models or document format constraints.
list_assets has no pagination support. If market grows to 100+ assets, response will bloat context window. Should add limit/offset params and return total_count.