Loading ToolBench
中国硬科技供应链信息差 MCP Server — Cloudflare Workers remote version. Provides structured access to signal boards, earnings tracking, and article listings for Chinese tech supply chain intelligence.
Server has 4 tools with complete input schemas and descriptions. Naming follows verb_noun pattern (read_*, get_, ask_). Descriptions are substantive (100-200 chars) and explain the domain context well. However, output schemas are not formally documented, responses are JSON strings without declared field structures. Parameters lack enums for constrained values (e.g., 'result' and 'status' accept free-form strings instead of enums). Error handling is minimal, no recovery guidance or actionable error messages. No tool annotations (readOnlyHint, etc.). Security: tools fetch from public GitHub Pages; no auth mechanism visible despite mentioning optional Bearer Token in comments.
问答式信息差查询：Agent 用自然语言问产业问题，返回最相关的信号+信息差依据。区别于数据 API——返回的是判断层（what it means），不是原始数据。
读取信号验证记录（track record）：命中率、已验证/待验证统计、每条信号从 predicted_on 到 verify_by 的结果。这是唯一用时间积累的资产——判断层的命中历史，LLM 无法生成。
读取中报信号验证追踪表：我们信号板预测的公司 vs 实际披露业绩 vs 验证状态（验证/部分验证/待验证）
读取信号信息差看板，返回结构化 JSON，包含 📊 高信息差信号（真信息差）。每条信号含 predicted_on/verify_by/verify_event/result 验证记录字段。
Output schemas not documented. Tools return JSON strings without declared field structures (e.g., read_signal_board returns {total, signals, source} but this is not in inputSchema or a separate outputSchema). LLMs cannot plan downstream calls or extract fields reliably.
Parameters accept free-form strings instead of enums. 'result' param in read_signal_board accepts '待验证'/'已兑现'/'部分兑现'/'落空'/'推迟' but is declared as z.string().optional() with no enum constraint. LLMs may hallucinate invalid values.
No error handling or recovery guidance. If data source fails (404, timeout), tools throw generic errors with no actionable next steps for the LLM. No categorization of retryable vs fatal errors.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | 2026-07-28+ | v2 |
No tool annotations. Tools lack readOnlyHint, destructiveHint, or idempotentHint. All tools are read-only but this is not declared in the schema, forcing LLMs to infer safety from descriptions.
Bearer Token auth mentioned in comments but not enforced in code. MCP_ACCESS_TOKEN env var is checked but no actual validation logic visible. Public data source (GitHub Pages) may not require auth, but the incomplete implementation creates security ambiguity.