---
name: people-search-exa
description: People search using Exa. Finds public profiles across the web (social + professional) and returns structured results with sources.
triggers:
- people search
- find a person
- find experts
- find candidates
- recruiting research
- sales prospecting
- find social profiles
- person lookup
requires_mcp: exa
context: fork
---
# People Search (Exa)
## Tool Restriction (Critical)
ONLY use `people_search_exa`. Do NOT use other Exa tools.
Note: If the user asks for `linkedin_search_exa`, use `people_search_exa` instead—it covers LinkedIn and other social/professional profiles.
## Token Isolation (Critical)
Never run Exa searches in main context. Always spawn Task agents:
- Agent calls `people_search_exa`
- Agent merges + deduplicates results (by URL AND near-duplicates) before presenting
- Agent extracts only what's needed (name, role, company, location, profile URLs, key evidence)
- Agent returns distilled output (brief markdown or compact JSON)
- Main context stays clean regardless of search volume
## When to Use
Use `people_search_exa` when you need **broad people discovery** across the web, such as:
- Discover people by role + company + region (sales/recruiting)
- Find experts by topic (e.g., "incident response consultants specializing in AWS")
- Build shortlists of candidates/prospects with certain skills
If the user needs general web sources or long-form articles, use a web search tool instead.
## Inputs (Supported)
`people_search_exa` supports:
- `query` (string, required)
- `numResults` (number, optional)
## Query Hygiene (Improves Precision)
People Search works best with **attribute-rich** queries. Encourage the user (or infer) these attributes:
- Role/title + seniority (e.g., "Staff", "Director", "VP")
- Company (current or prior)
- Region/time zone/market (optional but helpful)
- Skills/stack/domain keywords
- Industry (optional)
Examples:
- "enterprise account executives from Microsoft in EMEA"
- "VP of Product at Microsoft"
- "Rust compiler engineers Europe"
- "cloud security consultants incident response AWS"
## Dynamic Tuning
Avoid hardcoding result counts. Tune dynamically:
- Find a single known person → `numResults` 3–8
- "a few" → `numResults` 10–20
- "comprehensive" / shortlist building → `numResults` 20–50, then post-filter
## Query Variation
Generate 2–3 query variations for coverage, run in parallel, then merge + dedupe.
Good variation strategies:
- title synonyms (VP/Head/Director)
- adding/removing location
- swapping skill keywords (e.g., "GenAI" ↔ "LLMs")
## Output Format (Recommended)
Return:
1) Results (structured list; one person per row)
2) Sources (URLs; 1-line relevance each)
3) Notes (uncertainty/conflicts)
Before presenting:
- Deduplicate similar results (mirrors/reposts/profile variants) and keep the best representative source per person.
## Browser Fallback
Auto-fallback to browser automation when:
- Exa results are insufficient for identification
- A target page is auth-gated / dynamic
- User needs verification from a specific page that Exa can't surface cleanly
## MCP Configuration
```json
{
"servers": {
"exa": {
"type": "http",
"url": "https://mcp.exa.ai/mcp?tools=people_search_exa"
}
}
}
```
