exa-search

name	exa-search
description	Search the web with the local Exa CLI. Use for Exa-powered research, recent sources, domain-filtered search, similar-page discovery, URL content extraction, or cited web answers.

Exa Search

Use the local exa-cli installation to search the web, inspect URLs, and produce grounded answers.

Quick Start

Prefer exa-cli from PATH. If it is missing, check /Users/melvynx/.local/bin/exa-cli. The local project source is /Users/melvynx/.cli/exa-cli.

Verify availability and auth before doing substantive work:

exa-cli --help
exa-cli auth test

If authentication is missing or invalid:

exa-cli auth set <token>
exa-cli auth test

Workflow

1. Choose the smallest command that matches the task.
2. Use --json when calling the CLI programmatically.
3. Start with concise output, then add --text, --summary, or --highlights only when needed.
4. Use domain, date, and category filters when freshness or source quality matters.
5. Summarize findings with the returned URLs so the final answer stays grounded.

Command Selection

Use exa-cli search query for normal web research and source discovery.

exa-cli search query --query "latest AI research" --json
exa-cli search query --query "best React frameworks" --type neural --num-results 5 --json
exa-cli search query --query "SpaceX news" --start-date 2024-01-01 --category news --summary --json

Use exa-cli search find-similar when you already have one strong URL and want neighboring sources.

exa-cli search find-similar --url "https://example.com" --num-results 5 --json

Use exa-cli contents get when you already know the URL or URLs and need the page content.

exa-cli contents get --urls "https://example.com" --text --json
exa-cli contents get --urls "https://a.com,https://b.com" --summary --json

Use exa-cli answer query when the user wants a direct answer grounded in web citations.

exa-cli answer query --query "What is the latest valuation of SpaceX?" --json
exa-cli answer query --query "Who won the 2024 election?" --text --json

Use exa-cli context get when you need compact research or code context under a token budget.

exa-cli context get --query "how to use React hooks" --tokens-num 4000 --json
exa-cli context get --query "Python async await patterns" --tokens-num 8000 --json

Query Guidance

Use the user's full intent as the query whenever possible. Avoid single-word queries unless the task is intentionally broad.

Prefer these patterns:

• Current events or recent changes: add --start-date, --end-date, or a date in the query.
• Source curation: add --include-domains or --exclude-domains.
• Narrowing intent: use --category for news, company, research_paper, pdf, or tweet.
• Deep inspection: add --text, --summary, or --highlights only after you have the right result set.

Common Patterns

Domain-filtered search:

exa-cli search query --query "React server components caching" --include-domains react.dev,nextjs.org --json

Recent news scan:

exa-cli search query --query "OpenAI funding" --category news --start-date 2026-03-01 --num-results 5 --json

Follow-up content fetch after search:

exa-cli contents get --urls "https://example.com/post-1,https://example.com/post-2" --text --json

Grounded answer first, manual inspection second:

exa-cli answer query --query "What changed in Bun's package manager recently?" --json

Error Handling

If exa-cli auth test fails, tell the user the Exa token is missing or invalid and ask them to authenticate with exa-cli auth set.

If the CLI is missing from PATH, try /Users/melvynx/.local/bin/exa-cli before assuming it is not installed.

If a query returns weak results:

1. Rewrite the query with more concrete intent.
2. Add filters for domain, category, or date.
3. Switch from answer query to search query plus contents get when you need tighter control over sources.

Common Mistakes

• Do not start with --text on broad searches unless you actually need page bodies.
• Do not use vague one-word queries when a full natural-language query is available.
• Do not rely on answer query alone for high-stakes topics if you should inspect the cited URLs directly.
• Do not fetch contents for many URLs before narrowing the result set with search.