MCP server for SearXNG metasearch engine integration. Provides web search and URL reading capabilities via FastMCP framework.
The server defines two tools with complete input schemas and reasonable descriptions. Both tools have proper parameter types and descriptions, and include tool annotations (readOnlyHint, openWorldHint). However, there are notable gaps: output schemas are not documented, error handling guidance is minimal, and some parameter descriptions could be more prescriptive about constraints and formats. The server validates environment variables at startup but lacks per-call error recovery guidance that LLMs need to self-correct. Tool naming is clear and follows verb_noun convention (searxng_web_search, web_url_read). The second tool's description is strong (mentions markdown conversion, pagination modes), while the first is adequate but generic.
Performs a web search using the SearXNG API. Use this to find information on the web. Aggregates results from multiple search engines.
Reads a URL and converts HTML/XML content into Markdown. Supports character pagination, section extraction, paragraph ranges, and heading-only mode.
No documented output schema. The tool definitions lack explicit documentation of return value structure. LLMs cannot plan downstream chaining or extraction without knowing what fields to expect.
Minimal error recovery guidance in descriptions. Error handling exists in code (e.g., timeout, invalid URI scheme), but tool descriptions don't explain to LLMs what to do on failure: 'If timeout occurs, try again with a longer timeoutMs value' or 'URL scheme unsupported; supported schemes: http, https, ftp, sftp, smb, gemini, ipfs'.
Parameter constraints not fully described. The 'pageno' parameter (default: 1) lacks min/max bounds; 'max_results' (default: 10) lacks explicit upper limit. For web_url_read, 'timeoutMs' (default: 10000) has no stated range. LLMs may pass invalid values without constraint guidance.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 59 | <=2025-11-25 | v2 |
| 2026-03-09 | D | 58 | - | v1 |
searxng_web_search description is generic. 'Performs a web search using the SearXNG API' doesn't explain when to use this vs. a general web search tool, or what 'Aggregates results from multiple search engines' adds. Better: 'Search the web via SearXNG, aggregating results from multiple search engines (Google, Bing, DuckDuckGo, etc.). Use for live, multi-source queries. Returns title, URL, snippet, and relevance score.'
No explicit documentation of return types. Code shows format_search_result returns plain text with 'Title:', 'URL:', 'Content:', 'Score:' lines, and web_url_read returns markdown. But tool descriptions don't state this. LLMs need to know: 'Returns plain-text results with fields: Title, URL, Content (snippet), Score. One result per section.' or 'Returns markdown-formatted content with optional heading extraction and paragraph pagination.'