Loading ToolBench
MCP server for Perplexity AI search integration
Single tool 'perplexity_search' has a reasonable name and adequate description, but the parameter descriptions are minimal and lack context. The schema is properly structured with type constraints (enum, min/max), but the parameter descriptions are generic placeholders that don't guide LLM usage. Output schema is implicit (returns text content) but not documented. Error handling exists but is rudimentary, errors are returned as text without recovery guidance or categorization. The tool itself is well-formed and functional, but falls short of production-grade quality for agentic use.
Search using Perplexity AI's models with context-aware responses and citations
Parameter descriptions lack context and recovery guidance. 'Search query' and 'Model to use' are bare statements; they should explain what constitutes a valid query, when to use each model variant, and what the expected output will be.
Output schema is undocumented. The tool returns {content: [{type: 'text', text: string}], isError?: boolean}, but this structure is not declared anywhere. LLMs must infer the response format from code, error-prone and wastes tokens parsing unstructured results.
Error responses are unstructured text without recovery guidance. When an error occurs, the tool returns 'Error performing search: <message>' with isError=true. The LLM does not know if the error is retryable, whether to try a different model, or whether to ask the user for a better query.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 55 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 45 | - | v1 |
The 'count' parameter description is misleading. It says 'Number of results' but the implementation uses count*100 as max_tokens, which controls response verbosity, not result count. The parameter is semantically disconnected from its implementation.
Missing dependency hints and use-case guidance. The tool description does not explain when to use sonar-reasoning-pro vs sonar-pro, or when to call this tool vs alternative search strategies. Multi-step planning requires clearer intent signals.