---
name: web-search-advanced-exa
description: Advanced web search using Exa. Full control over filtering (domains, dates, category, text) plus highlights/summaries and subpage crawling.
triggers:
- advanced web search
- filtered web search
- domain filter
- date filter
- site filter
- search with highlights
- search with summary
- crawl subpages
requires_mcp: exa
context: fork
---
# Web Search Advanced (Exa)
## Tool Restriction (Critical)
ONLY use `web_search_advanced_exa`. Do NOT use `web_search_exa` or other Exa tools.
## Token Isolation (Critical)
Never run Exa searches in main context. Always spawn Task agents:
- Agent calls `web_search_advanced_exa`
- Agent merges + deduplicates results (by URL AND near-duplicates like syndicated/mirrored content) before presenting
- Agent extracts only what's needed for the user's request (facts, quotes, code snippets, dates)
- Agent returns distilled output (brief markdown or compact JSON)
- Main context stays clean regardless of search volume
## When to Use
Use `web_search_advanced_exa` when you need **control**:
- Restrict to specific domains (docs sites, arXiv, GitHub, etc.)
- Filter by published date or crawl date
- Force a category (news / research paper / github / pdf / etc.)
- Require/avoid certain words on the page (includeText/excludeText)
- Generate summaries or highlights for retrieval workflows
- Crawl subpages (e.g., pricing/docs/careers linked from a homepage)
If you just want a simple "search + clean context", prefer the basic tool (outside this skill).
## Inputs (Supported)
### Core search parameters
- `query` (string, required): Search query (question, statement, or keywords)
- `numResults` (number, optional; default: 10; range: 1–100)
- `type` (string enum, optional; default: "auto"):
- "auto", "fast", "deep", "neural"
### Category + domain filtering
- `category` (string enum, optional):
- "company", "research paper", "news", "pdf", "github", "tweet", "personal site", "people", "financial report"
- `includeDomains` (string[], optional): only return results from these domains
- `excludeDomains` (string[], optional): exclude results from these domains
Important: the Exa Search API notes that the `company` and `people` categories support a limited set of filters; using unsupported filters can error. For `people`, `includeDomains` is limited to LinkedIn domains.
### Date filtering (ISO 8601)
- `startPublishedDate` (string, optional): include results published after this date
- `endPublishedDate` (string, optional): include results published before this date
- `startCrawlDate` (string, optional): include results crawled after this date
- `endCrawlDate` (string, optional): include results crawled before this date
### Text filtering
- `includeText` (string[], optional): only include results containing ALL of these strings
- `excludeText` (string[], optional): exclude results containing ANY of these strings
### Content extraction + generation
- `textMaxCharacters` (number, optional): cap text extracted per result
- `contextMaxCharacters` (number, optional; default: 10000): cap the aggregated context string
- `enableSummary` (boolean, optional; default: false): generate a short summary for each result
- `summaryQuery` (string, optional): focus the summary on a specific question
- `enableHighlights` (boolean, optional; default: false): extract highlight snippets
- `highlightsNumSentences` (number, optional): sentences per highlight
- `highlightsPerUrl` (number, optional): number of highlights per URL
- `highlightsQuery` (string, optional): focus highlights on a specific question
### Live crawl + subpages
- `livecrawl` (string enum, optional; default: "fallback"):
- "never", "fallback", "always", "preferred"
- `livecrawlTimeout` (number, optional; default: 30000): milliseconds
- `subpages` (number, optional): crawl N subpages per result (typically 1–10)
- `subpageTarget` (string[], optional): keywords to guide which subpages to pick (e.g., ["pricing", "docs", "careers"])
### Additional controls
- `userLocation` (string, optional): two-letter ISO country code (e.g., "US")
- `moderation` (boolean, optional): filter unsafe content
- `additionalQueries` (string[], optional): extra query variations (best paired with `type: "deep"`)
## Dynamic Tuning
Avoid hardcoding. Tune parameters to intent:
- Quick factual lookup → `type: "fast"`, `numResults: 3–8`
- High recall / "don't miss anything" → `type: "deep"`, `numResults: 20–50`, plus `additionalQueries`
- Best semantic match → `type: "neural"` or `"auto"`
- "Latest/current/today" → set `livecrawl: "preferred"` (or `"always"` if freshness is critical)
Use `contextMaxCharacters` to control how much aggregated context comes back.
Use `textMaxCharacters` to control per-result payload size.
## Query Variation
You have two options:
1) Multiple calls (2–3 query rewrites), then merge + dedupe
2) Single call using `additionalQueries` to expand coverage
### Programming query hygiene (reduces noise)
If the query is programming-related, always include the programming language (and optionally framework/version)
to reduce cross-language noise.
Example: use **"Go generics"** instead of just **"generics"**.
## Examples
Domain-restricted search (docs-only):
```
web_search_advanced_exa {
"query": "rate limits and pricing",
"includeDomains": ["exa.ai"],
"type": "auto",
"numResults": 5
}
```
News within a published date window:
```
web_search_advanced_exa {
"query": "AI agent security incidents",
"category": "news",
"startPublishedDate": "2025-12-01T00:00:00.000Z",
"endPublishedDate": "2026-01-27T00:00:00.000Z",
"numResults": 10,
"type": "auto"
}
```
Exclude noisy results and focus on a term:
```
web_search_advanced_exa {
"query": "startup pricing page",
"includeText": ["pricing"],
"excludeText": ["course"],
"numResults": 8,
"type": "fast"
}
```
Enable highlights + summaries for RAG-style extraction:
```
web_search_advanced_exa {
"query": "MCP protocol best practices",
"numResults": 8,
"type": "deep",
"enableHighlights": true,
"highlightsPerUrl": 2,
"highlightsNumSentences": 2,
"enableSummary": true,
"summaryQuery": "What are the key best practices and pitfalls?"
}
```
Deep search with query expansion + subpage crawling:
```
web_search_advanced_exa {
"query": "Company X pricing and enterprise plan",
"type": "deep",
"additionalQueries": ["Company X pricing", "Company X enterprise plan", "Company X billing"],
"numResults": 5,
"subpages": 3,
"subpageTarget": ["pricing", "plans", "billing", "enterprise"],
"livecrawl": "preferred"
}
```
## Output Format (Recommended)
Return:
1) Answer (short, structured)
2) Sources (URLs, 1-line relevance each; include dates if relevant)
3) "What's uncertain / conflicting" if needed
Before presenting:
- Deduplicate similar results (mirrors/syndication/reposts) and keep the best representative source per claim.
## MCP Configuration
```json
{
"servers": {
"exa": {
"type": "http",
"url": "https://mcp.exa.ai/mcp?tools=web_search_advanced_exa"
}
}
}
```
