home / skills / plurigrid / asi / exa-search

exa-search skill

safe

This skill helps you perform semantic web searches with Exa to retrieve contextual high quality documentation and research results.

npx playbooks add skill plurigrid/asi --skill exa-search

Review the files below or copy the command above to add this skill to your agents.

Files (1)

SKILL.md

2.0 KB

---
name: exa-search
description: Use Exa for semantic/neural web search. Exa understands context and returns high-quality results. Use this skill when you need to search the web for documentation, research, or any information that requires understanding meaning rather than just keyword matching. NEVER substitute web_search for Exa - they serve completely different purposes.
---

# Exa Semantic Search

Exa provides neural/semantic search via MCP. Use it for high-quality web search that understands context.

## When to Use Exa

- Searching for documentation or technical information
- Research requiring semantic understanding
- Finding information where exact keywords are unknown
- Company research and LinkedIn searches
- Deep research tasks

## When NOT to Use Exa

- Never use `web_search` as a substitute - it's basic keyword matching only
- If Exa fails, troubleshoot Exa - don't fall back to `web_search`

## Available Tools

The Exa MCP server provides these tools:

- `web_search_exa` - Semantic web search
- `crawling_exa` - Crawl and extract web content
- `company_research_exa` - Research companies
- `linkedin_search_exa` - Search LinkedIn profiles
- `deep_researcher_start` - Start deep research task
- `deep_researcher_check` - Check deep research status

## Configuration

Exa is configured as a remote HTTP MCP in `~/.mcp.json`:

```json
{
  "exa": {
    "type": "http",
    "url": "https://mcp.exa.ai/mcp?tools=web_search_exa,crawling_exa,company_research_exa,linkedin_search_exa,deep_researcher_start,deep_researcher_check"
  }
}
```

## Usage Examples

### Basic Search
Use the Exa MCP tools directly when semantic search is needed.

### Deep Research
1. Start with `deep_researcher_start` for complex topics
2. Poll with `deep_researcher_check` until complete
3. Get comprehensive, synthesized results

## Critical Rules

1. **NEVER replace Exa with web_search** - they are fundamentally different
2. **NEVER use web_search in Task sub-agents** as a substitute for Exa
3. If Exa fails, troubleshoot Exa - do not substitute

Overview

This skill provides access to Exa, a neural semantic web search service that understands context and returns high-quality, meaning-aware results. Use it when keyword matching is insufficient and you need search results that capture intent, relationships, and nuanced relevance. It is optimized for documentation, technical research, company due diligence, LinkedIn discovery, and deep investigative workflows.

How this skill works

The skill calls Exa MCP tools over HTTP to perform semantic search, site crawling, company research, LinkedIn profile discovery, and managed deep research tasks. Queries are processed using neural retrieval so results are ranked by meaning and context rather than simple keyword overlap. For complex investigations you can start a deep research job and poll its status until it returns a comprehensive synthesis.

When to use it

Looking up technical documentation where exact terms are unknown or inconsistent
Research tasks that require understanding relationships, intent, or synonyms
Company analysis, competitor research, or LinkedIn profile discovery
Deep research projects that need staged, synthesized results
Crawling and extracting content from websites for semantic indexing

Best practices

Always prefer Exa for semantic needs; do not substitute basic keyword web_search
Frame queries with context and desired scope (time range, domain, or document type)
Use deep_researcher_start for complex multi-step investigations and poll with deep_researcher_check
If Exa returns unexpected results, troubleshoot Exa configuration or query formulation before switching tools
Combine crawling_exa with web_search_exa for targeted site extraction followed by semantic ranking

Example use cases

Find authoritative API documentation and examples when endpoint names vary across versions
Conduct competitor or investor research by combining company_research_exa with LinkedIn profile data
Start a deep research job on a scientific topic and retrieve a synthesized literature overview
Crawl a vendor site to extract product specs and then semantically search the extracted content
Discover relevant developer discussions and threads where keywords differ from formal terminology

FAQ

Can I use web_search instead of Exa for semantic queries?

No. web_search is keyword-based and should never substitute Exa for semantic needs.

What if Exa fails to return useful results?

Troubleshoot Exa configuration or refine your query context. Use deep_researcher tools for complex tasks rather than falling back to web_search.