MCP server that provides internet search capabilities using SearXNG, a privacy-focused metasearch engine.
Single tool 'search' has a well-structured Zod schema with proper types and descriptions for all parameters. Tool description is adequate (67 chars) but generic. Parameter descriptions are present and specific (e.g., 'Comma separated list of active search categories'). Output schema is not explicitly documented, the response structure is inferred from formatResultsForLLM() but not formally declared. Error handling returns text responses but lacks actionable recovery guidance. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite READ_ONLY risk classification.
Search the internet using a variety of search engines.
Output schema not documented. Response structure (formatted text vs raw JSON) is determined by raw_json parameter, but the LLM cannot see what fields/structure to expect. This forces the LLM to parse unstructured text or guess JSON structure.
Error handling lacks recovery guidance. Errors return plain text like 'Error searching SearXNG: <error>' without actionable next steps. LLM cannot determine if the error is retryable, user-fixable, or fatal.
Tool description is generic (67 chars). Does not explain WHEN to use this tool vs other search tools, what prerequisites exist (SearXNG server must be running), or what the formatted output contains. Baseline for A+ tools is 50-200 chars with clear context.
No tool annotations despite READ_ONLY risk classification. Tool should declare readOnlyHint: true in the schema to signal to the LLM that this is a safe, non-destructive operation.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
Result limit not enforced or documented. formatResultsForLLM() returns all results without pagination or a cap. Large result sets (100+ items) will bloat the context window and degrade LLM reasoning.