Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Neural search via Exa MCP for web, code, and company research. Use when the user needs web search, code examples, company intel, people lookup, or AI-powered deep research with Exa's neural search engine.
.claude/skills/loulanyue-exa-search/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | -54% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 354% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 59% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 39% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 43% | 0% |
Neural search for web content, code, companies, and people via the Exa MCP server.
Exa MCP server must be configured. Add to ~/.claude.json:
json"exa-web-search": { "command": "npx", "args": ["-y", "exa-mcp-server"], "env": { "EXA_API_KEY": "YOUR_EXA_API_KEY_HERE" } }
Get an API key at exa.ai. This repo's current Exa setup documents the tool surface exposed here: web_search_exa and get_code_context_exa. If your Exa server exposes additional tools, verify their exact names before depending on them in docs or prompts.
General web search for current information, news, or facts.
web_search_exa(query: "latest AI developments 2026", numResults: 5)Parameters:
| Param | Type | Default | Notes | |-------|------|---------|-------| | query | string | required | Search query | | numResults | number | 8 | Number of results | | type | string | auto | Search mode | | livecrawl | string | fallback | Prefer live crawling when needed | | category | string | none | Optional focus such as company or research paper |
Find code examples and documentation from GitHub, Stack Overflow, and docs sites.
get_code_context_exa(query: "Python asyncio patterns", tokensNum: 3000)Parameters:
| Param | Type | Default | Notes | |-------|------|---------|-------| | query | string | required | Code or API search query | | tokensNum | number | 5000 | Content tokens (1000-50000) |
web_search_exa(query: "Node.js 22 new features", numResults: 3)get_code_context_exa(query: "Rust error handling patterns Result type", tokensNum: 3000)web_search_exa(query: "Vercel funding valuation 2026", numResults: 3, category: "company")
web_search_exa(query: "site:linkedin.com/in AI safety researchers Anthropic", numResults: 5)web_search_exa(query: "WebAssembly component model status and adoption", numResults: 5)
get_code_context_exa(query: "WebAssembly component model examples", tokensNum: 4000)web_search_exa for current information, company lookups, and broad discoverysite:, quoted phrases, and intitle: to narrow resultstokensNum (1000-2000) for focused code snippets, higher (5000+) for comprehensive contextget_code_context_exa when you need API usage or code examples rather than general web pagesdeep-research — Full research workflow using firecrawl + exa togethermarket-research — Business-oriented research with decision frameworks| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 17,923 | 7,989 | -55% | 1 | 1 | 0% | 3,475 | 1,214 | -65% | 0 | 0 | — |
case-12 | pass→pass | 6,119 | 1,744 | -71% | 1 | 1 | 0% | 1,064 | 1,208 | +14% | 0 | 0 | — |
case-02 | fail→fail | 12,735 | 7,145 | -44% | 1 | 1 | 0% | 2,393 | 1,477 | -38% | 0 | 0 | — |
case-03 | fail→fail | 14,821 | 6,469 | -56% | 1 | 1 | 0% | 2,297 | 1,330 | -42% | 0 | 0 | — |
case-04 | fail→fail | 3,203 | 4,468 | +39% | 1 | 1 | 0% | 616 | 1,831 | +197% | 0 | 0 | — |
case-05 | fail→fail | 30,676 | 7,374 | -76% | 1 | 1 | 0% | 6,173 | 1,577 | -74% | 0 | 0 | — |
case-06 | fail→fail | 9,379 | 4,875 | -48% | 1 | 1 | 0% | 1,672 | 1,218 | -27% | 0 | 0 | — |
case-07 | fail→fail | 4,480 | 7,128 | +59% | 1 | 1 | 0% | 734 | 2,091 | +185% | 0 | 0 | — |
case-08 | fail→fail | 28,391 | 6,774 | -76% | 1 | 1 | 0% | 4,964 | 1,348 | -73% | 0 | 0 | — |
case-09 | fail→pass | 14,814 | 3,773 | -75% | 1 | 1 | 0% | 2,599 | 1,200 | -54% | 0 | 0 | — |
case-10 | fail→pass | 2,139 | 1,658 | -22% | 1 | 1 | 0% | 275 | 1,249 | +354% | 0 | 0 | — |
case-11 | pass→pass | 3,639 | 1,824 | -50% | 1 | 1 | 0% | 565 | 1,225 | +117% | 0 | 0 | — |
case-13 | fail→pass | 4,768 | 2,141 | -55% | 1 | 1 | 0% | 810 | 1,291 | +59% | 0 | 0 | — |
case-14 | pass→pass | 9,904 | 2,171 | -78% | 1 | 1 | 0% | 1,953 | 1,255 | -36% | 0 | 0 | — |
case-15 | pass→pass | 3,226 | 1,997 | -38% | 1 | 1 | 0% | 474 | 1,281 | +170% | 0 | 0 | — |
case-16 | pass→pass | 5,028 | 3,415 | -32% | 1 | 1 | 0% | 861 | 1,501 | +74% | 0 | 0 | — |
case-17 | pass→pass | 3,143 | 2,984 | -5% | 1 | 1 | 0% | 490 | 1,401 | +186% | 0 | 0 | — |
case-18 | fail→pass | 6,976 | 4,020 | -42% | 1 | 1 | 0% | 1,178 | 1,638 | +39% | 0 | 0 | — |
case-19 | fail→pass | 6,763 | 3,507 | -48% | 1 | 1 | 0% | 1,102 | 1,578 | +43% | 0 | 0 | — |
case-20 | pass→pass | 14,252 | 7,964 | -44% | 1 | 1 | 0% | 2,512 | 2,265 | -10% | 0 | 0 | — |
case-21 | fail→fail | 38,125 | 7,903 | -79% | 1 | 1 | 0% | 6,182 | 1,416 | -77% | 0 | 0 | — |
case-22 | pass→pass | 8,660 | 7,704 | -11% | 1 | 1 | 0% | 1,574 | 2,371 | +51% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 15 counted toward the lift figure. The other 7 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +23 percentage points is the difference between those two pass rates over the 15 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.