Model Context Protocol server that integrates with Perplexity AI API to provide search, documentation, API discovery, and code deprecation checking capabilities
The server implements 4 tools with partial schema and description coverage. All tools have names that start with action verbs (search, get_, find_, check_) and non-empty descriptions. However, parameter descriptions are sparse or missing in several tools, and output schemas are not documented. The implementation shows basic tool structure but lacks the rigor expected for production quality. Tool definitions are visible in the source code, so no inference penalty applies. The schemas present are JSON Schema objects with required fields and enums, but lack depth in constraint documentation (e.g., no minLength/maxLength on strings, no range validation on numeric types). Error handling is basic (no recovery guidance, no actionable error messages visible in the tools). The similarity cache integration suggests some thought to performance, but the tool definitions themselves do not reflect the patterns needed for reliable LLM interaction.
Check if code patterns, libraries, or APIs are deprecated and get recommendations for modern alternatives
Discover and find relevant APIs related to a specific topic or functionality
Retrieve comprehensive documentation and usage examples for a given topic or library
Perform a general search query to get comprehensive information on any topic
Parameter descriptions are incomplete or missing context. E.g., 'detail_level' lacks explanation of how brief/normal/detailed affect API behavior or response length. 'search_recency_filter' lacks guidance on when to use each option or its default behavior.
Output schemas are not documented in tool definitions. Agents cannot predict what fields to expect from responses (e.g., does search return citations, formatted_text, sources, or raw_response?). This forces agents to infer structure and risks parsing errors.
Error handling in tool execution (e.g., SearchTool.execute) returns generic anyhow::Result with no recovery guidance. A missing PERPLEXITY_API_KEY returns an error but no hint that the user should set the env var or contact an admin. No categorization of errors (retryable vs fatal).
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 59 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 19 | - | v1 |
Parameter 'language' in check_deprecated_code is optional but lacks guidance on default behavior if omitted. Does the tool auto-detect language from code? Does it assume Python or JavaScript? The description says 'Optional: The programming language' but gives no hint of fallback behavior.
Tools accept no pagination parameters (limit, offset, page, cursor). If a search returns hundreds of results, no mechanism exists to page through them. The tool description for 'search' makes no mention of result count limits or how to retrieve additional results.
No input validation examples or constraints visible in schema definitions. E.g., 'query' parameter has no minLength/maxLength. 'detail_level' enum is present but description does not explain behavior differences. No constraints on API key format or validation that it was provided.