---

name: exa-search description: Use Exa's /search and /contents APIs to perform internet-scale retrieval, filter by domain/date, and fetch cached page text, highlights, or summaries.

Exa Search Skill

Use this skill whenever you need live-ish web search results, filtered link discovery, or cached page contents via the Exa Search API.

Setup

1. Export your Exa API key.

   Copyexport EXA_API_KEY="sk-your-key"
   

2. (Optional) override the base URL via EXA_API_BASE (defaults to https://api.exa.ai).

No third-party dependencies are required; the helper script only uses Python's standard library.

Usage

The helper supports both /search and /contents:

Copy# Search (default command)
./scripts/search.py --query "latest AI news" --type fast --num-results 5
./scripts/search.py search --query "vector database benchmark" --include-domain arxiv.org --text
./scripts/search.py search --query "China CRM market" --start-published-date 2024-01-01 --end-published-date 2024-12-31 --table-limit 3 --raw out.json
./scripts/search.py search --query "best langgraph tutorials" --highlights 2 "key steps" --text

# Get contents for known URLs or IDs
echo 'https://docs.exa.ai/reference/search' | \
  xargs ./scripts/search.py contents --url --text --highlights 2
./scripts/search.py contents --url https://exa.ai/blog --text --text-max-chars 2000
./scripts/search.py contents --id https://docs.exa.ai/reference/search \
  --summary-query "Key takeaways" --table-limit 1
./scripts/search.py contents --url https://exa.ai \
  --summary-schema examples/schema-company.json --text

Key search flags:

• --type: auto (default), fast, neural, deep.
• --num-results: up to 100 (subject to plan/type limits).
• --include-domain / --exclude-domain: repeatable filters.
• Date filters: --start-published-date, --end-published-date, --start-crawl-date, --end-crawl-date.
• --category: focus on news, research paper, github, people, etc.
• Content extraction: --text for cached body, --highlights [numSentences query] for snippets.
• Output: --table-limit N controls stdout preview, --raw file.json saves full API response.

Key contents flags:

• Targets: provide one or more --url or --id (IDs come from previous search results).
• --text, --text-max-chars, --text-include-html for body retrieval.
• --highlights [numSentences query], --highlights-per-url for excerpt control.
• --summary-query or --summary-schema <file> for free-form or structured summaries.
• --livecrawl controls freshness (never, fallback, preferred, always).

Workflow

1. Search: craft the query, tune filters, choose search type.
2. Inspect: review terminal output or saved JSON; note interesting result IDs/URLs.
3. Contents: pass those IDs/URLs to the contents subcommand to retrieve full text, highlights, or summaries.
4. Iterate: adjust highlights, livecrawl, or summary schema depending on downstream consumption (RAG, dashboards, etc.).

Reference

• Search reference: https://docs.exa.ai/reference/search
• Contents reference: https://docs.exa.ai/reference/get-contents
• Quickstart / SDK examples: https://docs.exa.ai/reference/quickstart

Troubleshooting

• 401 Unauthorized: ensure EXA_API_KEY is set and valid.
• Empty results: broaden query, remove restrictive filters, or try --type neural/deep.
• Slow responses: prefer cached content (--text), avoid live crawl when possible.
• Contents call errors: confirm each URL/ID exists and is crawlable; try --livecrawl fallback for stale links.