Broxy - Convert any webpage into API and MCP services. Browser extension that transforms web pages into APIs and MCP-compatible services.
Broxy Extension exposes 4 tools, all with minimal definition quality. Tool names lack action verbs (mcp_tools_list is borderline acceptable, but mcp_config, skill_md, and swagger are noun-only or unclear). Descriptions are present but extremely brief (10 - 28 characters), all under the 34-character p10 baseline. No input schemas are visible in the source code, all tools appear to be zero-parameter endpoint wrappers. Output schemas are entirely undocumented. The codebase shows route handlers but no formal parameter validation, type declarations, or LLM-optimized descriptions. Tools appear to be infrastructure/introspection utilities (listing tools, fetching config) rather than domain-focused capabilities. This is typical of extension scaffolding, not production agent tools.
返回 MCP 服务配置
返回所有MCP工具
返回 SKILL.md 文件内容
返回 OpenAPI/Swagger 格式的 API 文档
No input schemas defined for any tool. All tools appear to accept zero parameters, but this is inferred from absence of schema, not explicitly declared. Per HARD SCORING RULE, schema score for all tools MUST be 0.
Descriptions are extremely brief (10 - 28 chars), all below the 34-char p10 baseline. Examples: 'mcp_config' = '返回 MCP 服务配置' (18 chars), 'mcp_tools_list' = '返回所有MCP工具' (10 chars). Per HARD SCORING RULE, descriptions under 20 chars cannot exceed score 40, and most are under 20. These descriptions provide no context for LLM selection, no WHEN/WHY/WHAT detail.
Tool names do not follow verb_noun convention. 'mcp_tools_list' starts with a noun ('mcp'), not a verb. 'mcp_config', 'skill_md', 'swagger' are pure nouns. LLMs expect names like 'list_mcp_tools', 'get_mcp_config', 'get_skill_markdown'. Noun-only names reduce discoverability and clarity.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 28 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 27 | - | v1 |
No output schemas documented. Tools return unstructured responses (e.g., { tools }, { status, headers, body }) with no type declarations or field descriptions. LLMs cannot plan downstream tool calls without knowing the response shape.
Descriptions are in Chinese, limiting accessibility to non-Chinese-speaking developers and LLMs trained primarily on English. Industry standard for tool definitions is English.
No error handling guidance. Handlers return generic errors like { error: 'Route not found', ... } with no recovery suggestions or categorization. LLMs cannot determine if an error is retryable, user-fixable, or fatal.