Loading ToolBench
This MCP server exposes 3 tools for Archery SQL audit platform integration. All tools are properly registered with FastMCP and have descriptions. However, significant issues undermine production readiness: (1) Chinese-language descriptions create comprehension barriers for English-speaking LLM agents; (2) parameter descriptions exist but are sparse and lack format/constraint guidance; (3) no output schemas are documented, LLMs cannot predict return structure; (4) no error handling guidance; (5) tool composition is reasonable but authentication flow could be better guided. The server does follow basic verb naming conventions (get_*, refresh_*), and parameters are typed. Average per-tool score is 58.
认证 Archery 平台，获取 JWT Token。 如果已通过环境变量 ARCHERY_USERNAME / ARCHERY_PASSWORD 配置了凭证， 可以不传参数直接调用。否则需要提供 username 和 password。
获取 Archery SQL 上线工单清单。 支持按多种条件过滤，所有筛选参数均为可选。 传入 workflow_id 可查询单个工单的详细信息（包含 SQL 内容）。
手动刷新 Archery 的 access token。 通常不需要手动调用——当 token 过期时系统会自动刷新。 仅在需要强制刷新时使用。
Descriptions are in Mandarin Chinese exclusively, making them inaccessible to English-speaking LLM agents and reducing discoverability in English-centric tool ecosystems.
No output schemas documented for any tool. LLMs cannot predict return types, field names, or structure. This blocks tool chaining and forces LLMs to inspect responses blindly, increasing errors and wasting tokens.
get_workflow accepts workflow_status as free-form string but documents valid enum values only in the description. Should declare as JSON Schema enum to prevent hallucinated invalid values (e.g., 'workflow_pending', 'pending', 'queued').
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 40 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 47 | - | v1 |
No pagination guidance for get_workflow. Response likely returns large lists; description does not state default/max page_size, whether total count is returned, or how to handle result limits. LLMs cannot predict if 50 workflows will be returned or 5000.
Parameter descriptions lack format guidance. Date parameters (create_time_gte, create_time_lt) state 'ISO 格式' but do not specify whether timezone is required (e.g., UTC vs local). Numeric parameters (page, page_size) lack min/max bounds.
No error handling guidance. Tools raise ValueError and rely on client to handle. No recovery suggestions provided to LLM (e.g., 'If authentication fails, check ARCHERY_BASE_URL is set' or 'If token refresh fails, call get_auth_token again').
Authentication flow is not explicitly guided. LLM does not know it must call get_auth_token before get_workflow, or that environment variables (ARCHERY_USERNAME, ARCHERY_PASSWORD) can satisfy get_auth_token without parameters.