Exa Web Search (Free): Neural AI Search — No API Key, Three Specialized Search Modes

12,400+ downloads and 41 stars — the Exa Web Search (Free) Skill by @Whiteknight07 brings Exa's neural search capabilities to Clawdbot with zero API key setup. Via the mcporter MCP runner, the skill provides three specialized search modes: web search, code context search, and company research — all using Exa's free public tier.

What is Exa?

Exa (formerly Metaphor) is an AI-native search engine built end-to-end on transformer models — the first web-scale search engine that doesn't use keyword matching. Instead of comparing query terms against document keywords (the 50-year-old approach), Exa uses "next-link prediction": every indexed page is encoded as a semantic vector, and search works by predicting which links a given query would most plausibly point to. This makes it exceptionally strong at finding content that's conceptually related to a query, even without exact keyword overlap.

Exa is purpose-built for AI agents and LLM pipelines — results come back as clean structured data, not HTML full of ads and navigation chrome.

The zero-config free tier: On February 2, 2026, Exa opened unauthenticated access to their public MCP endpoint at mcp.exa.ai/mcp. This is what the skill exploits — no API key, no signup, immediate access. Rate limits apply to unauthenticated usage; heavy users can register for a standard free tier ($10 credits, no expiration, no credit card, ~2,000 searches at $5/1,000 queries).

Setup

# Install mcporter (MCP runner)
# (install instructions depend on your system)
 
# Verify exa is configured
mcporter list exa
 
# If not listed, add the Exa MCP server
mcporter config add exa https://mcp.exa.ai/mcp
 
# That's it — no API key needed

Three Core Search Modes

1. Web Search — Current News and General Info

# Latest news
mcporter call 'exa.web_search_exa(query: "latest AI breakthroughs 2026", numResults: 5)'
 
# Research topics (deep mode)
mcporter call 'exa.web_search_exa(query: "how does RAG work in LLMs", type: "deep", numResults: 8)'
 
# Fast lookup
mcporter call 'exa.web_search_exa(query: "quantum computing news", type: "fast")'

Search type parameter:

2. Code Context Search — GitHub and Stack Overflow

# Language fundamentals
mcporter call 'exa.get_code_context_exa(query: "Python asyncio basics and examples", tokensNum: 3000)'
 
# Framework patterns
mcporter call 'exa.get_code_context_exa(query: "Next.js 14 app router authentication middleware")'
 
# Debugging help
mcporter call 'exa.get_code_context_exa(query: "fixing CORS errors in Node.js Express")'
 
# Full API documentation
mcporter call 'exa.get_code_context_exa(query: "Stripe checkout session implementation", tokensNum: 5000)'

tokensNum guidance:

• 1000–2000 — Focused, specific examples
• 3000–5000 — Standard documentation lookup (default: 5000)
• 5000–50000 — Deep dives, full API docs

3. Company Research — Business Intelligence

# Startup research
mcporter call 'exa.company_research_exa(companyName: "Anthropic", numResults: 3)'
 
# Competitive intelligence
mcporter call 'exa.company_research_exa(companyName: "Perplexity AI")'
 
# Investment research
mcporter call 'exa.company_research_exa(companyName: "OpenAI", numResults: 5)'

Advanced Tools (Off by Default)

The skill has a full configuration mode (exa-full) that unlocks additional tools:

# Deep research mode (comprehensive)
mcporter call 'exa-full.deep_search_exa(query: "AI safety alignment comprehensive overview")'
 
# Web search with domain filters
mcporter call 'exa-full.web_search_advanced_exa(query: "machine learning tutorials", includeDomains: ["github.com", "arxiv.org"])'
 
# Date-filtered search
mcporter call 'exa-full.web_search_advanced_exa(query: "AI developments", startPublishedDate: "2026-01-01")'
 
# Extract content from specific URL
mcporter call 'exa-full.crawling_exa(url: "https://anthropic.com/news/claude-3-5-sonnet")'
 
# People/professional profiles
mcporter call 'exa-full.people_search_exa(query: "Yann LeCun AI researcher")'
 
# Async deep research task
mcporter call 'exa-full.deep_researcher_start(topic: "quantum computing applications in cryptography", depth: "comprehensive")'
mcporter call 'exa-full.deep_researcher_check(taskId: "abc123")'

Exa vs. Traditional Search Skills

With Clawdbot

Search for the latest news on transformer architecture
Find code examples for implementing OAuth in Express
Research what Anthropic has been working on recently
Find React hooks examples with 3000 tokens of context

Clawdbot routes these to the appropriate Exa tool automatically.

Considerations

• Rate limits apply — the free public tier has usage limits; heavy users may hit rate caps without a paid API key
• Requires mcporter — the skill uses the MCP runner architecture; mcporter must be installed and configured first
• Neural search trade-offs — Exa's semantic approach excels at "find similar to" queries but may miss exact keyword matches that traditional search engines handle well
• Advanced tools off by default — the full config unlocks powerful features like deep researcher and URL crawling; enable when needed
• No direct integration with other skills — Exa results are returned as text; combine with other skills for downstream processing

The Bigger Picture

The Exa Web Search (Free) Skill represents a different philosophy from traditional search integration: start with neural understanding of query intent, specialize by content type (code vs. companies vs. general web), and make it available for free. For developers especially, the code context search tool — pulling from GitHub and Stack Overflow with semantic matching — is genuinely more useful than a Google search for many coding questions. With 12,400+ downloads, the ClawHub community has validated this approach.

View the skill on ClawHub: exa-web-search-free