Loading ToolBench
The server defines one tool (perplexity_search_web) with a complete JSON Schema and basic descriptions. However, there are significant gaps in parameter documentation, output schema documentation, and error handling guidance. The tool name follows verb_noun convention and the input schema includes type definitions and an enum constraint for the 'recency' parameter. The description is present but generic (79 chars), lacking actionable guidance on when to use the tool or what it returns. Parameter descriptions are minimal: 'query' has a 46-char description but 'recency' has a lengthy one (211 chars) describing the enum values. The tool lacks documented output schema, error recovery guidance, and constraint-validation details. The prompts feature is implemented but mirrors the tool functionality with minimal additional value.
Search the web using Perplexity AI with recency filtering
Output schema not documented. Tool returns TextContent with stringified result, but the structure, format (markdown/plaintext), and fields (citations, source URLs) are not formally specified. LLMs cannot plan downstream tool calls or extract structured data without knowing what fields to expect.
No error handling or recovery guidance. The code calls response.raise_for_status() but provides no error messages to guide LLM retry logic. Missing API key, malformed queries, rate limits, or network failures will surface raw exceptions without actionable guidance like 'API key invalid, check PERPLEXITY_API_KEY environment variable' or 'Rate limit reached, retry in 60s'.
Parameter 'recency' description is overly verbose (211 chars, exceeds baseline p90=392 for tool description but excessive for a single enum param). The enum values are duplicated in description and schema, describe the semantic meaning (recency filters) once in schema, reference it in description.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 57 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 46 | - | v1 |
No result limiting or pagination guidance. Tool returns all results from Perplexity API without documented limits. If the API returns verbose metadata or multiple citations, the response could exceed reasonable token budgets. Baseline pattern expects cap at 20-50 items and explicit limit in tool description.
Prompts feature mirrors tool but adds minimal value. Prompts return generic multi-message format repeating the query and recency filter as separate messages. Prompts should either be removed (tool is sufficient) or provide genuinely distinct use cases (e.g., advanced research prompt, citation-focused prompt).
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). The tool is read-only and idempotent, but these properties are not declared in the Tool schema. Tool annotations enable agents to reason about side effects and retry logic.