Company Research (Exa)

Tool

Call the native MCP tool exa-search__web_search_advanced_exa directly. Parameters go in a JSON object — examples below use shorthand key=value.

Critical Rules

• Never run searches in main context. Always spawn subagents via llm_task or run in sequence with isolated results.
• Query variation: generate 2-3 phrasings, run in parallel, merge + deduplicate.
• Dynamic numResults: "a few" → 10-20, "comprehensive" → 50-100, explicit number → match it.

Categories

Start with category=company for discovery, then use other categories for deeper research.

Category Restrictions

With category=company: do NOT use includeDomains, excludeDomains, or date filters — causes errors. includeText/excludeText: only single-value strings.

Call Examples

Tool: exa-search__web_search_advanced_exa

# Company discovery
{ query: "AI infrastructure startups San Francisco", category: "company", numResults: 20, type: "auto" }

# Deep dive on specific company
{ query: "Anthropic funding rounds valuation 2024", type: "neural", livecrawl: "fallback", numResults: 10,
  includeDomains: ["techcrunch.com","crunchbase.com","bloomberg.com"] }

# News coverage
{ query: "Anthropic AI safety 2024", category: "news", numResults: 15, startPublishedDate: "2024-01-01" }

# Tweets/social
{ query: "Anthropic Claude new features", category: "tweet", numResults: 20 }

Output Format

Return:

1. Results — structured list, one company per row (name, URL, description, key facts)
2. Sources — URLs with 1-line relevance
3. Notes — any uncertainty or conflicts

Browser Fallback

If Exa returns insufficient results or content is auth-gated → use browser tool as fallback.