web-search-via-searxng

SearXNG Web Search

Use the configured SearXNG MCP server to search the web.

When to Use

• Answering questions that need up-to-date information
• Looking up documentation or API references
• Researching technical topics
• Finding news or current events
• Searching for images, videos, academic papers, code, packages

How to Search

Prefer delegating to the web-searcher agent via the Agent tool. The agent handles multi-language parallel search, source credibility verification, cross-validation, iterative deepening, and structured summarization — returning a clean result without polluting your context window with raw JSON. Only call the MCP search tools directly when:

• The web-searcher agent is unavailable (e.g., no subagent support)
• You need a single quick lookup where spawning an agent is overkill

MCP Tools Reference

When searching directly (without the agent), use these tools:

search

Main search tool. Parameters:

• query (required): Search terms
• categories: Comma-separated category names (e.g., 'general,news,science'). Prefer this over engines — categories leverage multiple engines automatically
• language: Language code (e.g., zh, en, ja). Filters results to the specified language
• time_range: day, week, month, year
• safesearch: 0 (off), 1 (moderate), 2 (strict)
• pageno: Starting page number (default 1)
• pages: Number of pages to fetch in parallel, 1-5 (default 1). Use 2-3 for comprehensive research
• max_results: Maximum results to return (default 10, max 100)
• format: compact (default, title/url/content only) or full (includes engines, score, dates)
• engines: Comma-separated engine names (e.g., google,arxiv,wikipedia). Only use when you need a specific source

autocomplete

Get search query suggestions. Use before searching to discover relevant terms. Best results come from 1-2 meaningful keywords (e.g., "python async"). Single characters return overly broad suggestions; full sentences return none. Makes an external API call to the configured backend (e.g., Bing, Google).

• query (required): 1-2 keyword query to autocomplete (e.g., "python async", not "p" or full sentences)

engine_info

Discover available search engines and categories. No parameters. Returns engines grouped by category. Response cached for 5 minutes.

Use this when you need to target specific engines or categories (e.g., "search academic papers" → call engine_info to find science engines, then search with categories=science).

Category Selection Guide

Use categories as the primary filter. Only fall back to engines when categories produce poor results.

User intent	Category	When to use engines instead
Current events, breaking news	news	—
Academic papers, research	science or scientific publications	arxiv for preprints, pubmed for biomedical
Code, libraries, tech docs	it	github for repos, stackoverflow for Q&A
Package lookup (npm, pip)	packages	pypi, docker hub by name
Images, photos	images	—
Videos	videos	youtube for YouTube-specific
Maps, locations	map	—
Definitions, word meanings	dictionaries or define	—
Weather	weather	—
Translation	translate	—
Shopping, products	shopping	—
Torrents, file search	files	—
Social media posts	social media	—
General knowledge	general (default)	—

Rules

1. Delegate to web-searcher agent by default — it returns clean summaries and keeps raw data out of your context
2. Always include a Sources section at the end with clickable markdown links
3. Prefer categories over engines to narrow results — categories leverage multiple engines automatically
4. Only use engines when you need a specific source (e.g., arxiv for preprints, github for repos)
5. Use pages=3 when you need comprehensive results
6. Use language when the user writes in a specific language
7. Use format=full when you need to evaluate result quality (scores, engines)
8. Use autocomplete to refine ambiguous queries before searching
9. Use engine_info only when the category table above doesn't cover the use case