Find and profile companies using Exa's semantic search — discovery by category/stage/geography, competitor sets ("companies like X"), funding and news. Calls the Exa REST API through a local script; no MCP server needed, only an EXA_API_KEY.
Setup (once)
export EXA_API_KEY=your-key # or put EXA_API_KEY=... in a .env at the repo root
Get a key at https://dashboard.exa.ai/api-keys. The script auto-loads a .env from the repo root, so setup is one-time. Shared details: see exa-native-base.
Run it
python scripts/company_research.py "<describe the companies you want>" [-n N] [-c CATEGORY] [--json]
Examples:
python scripts/company_research.py "AI infrastructure startups in San Francisco" -n 15
python scripts/company_research.py "companies like Stripe" -n 10
python scripts/company_research.py "Anthropic funding rounds investors" -c news --start-published 2025-01-01
Output is a Markdown list (title · date · snippet) ending in sources_reviewed: N. Add --json for raw results, --text to include page text.
Token isolation (do this for anything non-trivial)
Don't run bulk searches in your main context. For multi-angle research, dispatch a subagent per workstream (e.g. one for competitors, one for funding, one for news), tell each to run the script and return a compact table, then merge + deduplicate. This keeps raw search output out of your context. See exa-native-base for the orchestration pattern.
Query patterns
category=company returns structured company data (description, funding, headcount). Write queries that describe the company, not just a name.
# By category / geo / stage
python scripts/company_research.py "category:company developer tools for API testing" -n 10
python scripts/company_research.py "Series B fintech payments companies" -n 10
# Competitive set (layer multiple angles, then dedupe)
python scripts/company_research.py "companies like Notion" -n 10
python scripts/company_research.py "collaborative docs software tools" -n 15
# Funding / investors (drop to general web or -c news for announcements)
python scripts/company_research.py "Mistral AI funding round raised investors" -c news -n 8
Category restriction: with category=company, avoid pairing includeText/date filters with domain filters — use a plain query for discovery, then switch to -c news or general web for time-bound digging.
Override categories with -c
company (default, structured profiles) · news (press/announcements) · linkedin profile (people at the company) · financial report (filings). For social chatter use the exa-x-search skill.
After you get results
- Treat results as similarity, not validation — skim titles/snippets and drop off-target rows before reporting.
- To deep-read a promising page:
python ../_shared/exa_client.py contents <url> --text. - To expand a competitor set from one good company:
python ../_shared/exa_client.py similar <url> -n 10. - Format the final answer as a table (name · URL · one-line description · key facts) + a short Sources list.