Loading ToolBench
MCP Server providing access to TianGong AI knowledge bases and search capabilities across multiple domains including education, ESG disclosures, patents, standards, textbooks, and more.
The server implements 10 search tools with consistent naming patterns and detailed parameter schemas. All tools follow a search_* verb convention which is clear and discoverable. Most parameters include type definitions and descriptions. However, output schemas are not documented in the source code, and error handling guidance is minimal. Tool descriptions are adequate (100-150 chars typical) but could be more LLM-optimized with clearer selection criteria and use-case boundaries.
Search ESG disclosures for relevant content.
Search the education knowledge graph for related concepts.
Search environmental education materials for relevant content.
Search EU Green Deal documents for relevant content.
Search the internal knowledge base for organization content.
Search patent abstracts and metadata for relevant inventions.
Search sustainability and policy reports for relevant content.
Output schemas not documented in source code. Tool descriptions state what is returned (e.g., 'chunk results'), but the exact field structure, types, and pagination info are not visible. LLMs cannot plan downstream operations without knowing returned fields.
Error handling not visible in tool definitions. No indication of what errors can occur, whether they are retryable, or what the LLM should do next (e.g., retry, refine query, ask user). This violates pattern:recovery-guide.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 14 | 2026-07-28+ | v2 |
| 2026-03-09 | C | 66 | - | v1 |
Search academic publications for scientific findings.
Search environmental standards for applicable guidance.
Search environmental textbooks for supporting material.
Filter and dateFilter parameters lack explicit examples or guidance on valid formats. dateFilter uses UNIX timestamps but no examples show whether seconds or milliseconds are expected. Descriptions state 'Use only when user explicitly requests' but LLMs often mis-parse filter structures.
Tool descriptions do not clarify use-case boundaries or selection criteria. All tools follow the same 'Search X for relevant content' pattern. LLMs cannot easily distinguish when to use Search_Edu_Tool vs Search_Textbook_Tool vs Search_Report_Tool vs Search_Sci_Tool, all appear equivalent.
topK and extK parameters lack constraints (no min/max). Large values could blow context windows or cause timeouts. Baselines suggest 20-50 item limits for search results.
metaContains parameter (in Search_ESG_Tool and Search_Standard_Tool) is described as 'fuzzy matching' but no explanation of matching behavior, case sensitivity, or substring rules provided. LLMs will guess wrong about whether 'env' matches 'environmental'.