AGENTIC MEMORY

Skills

Useful instructions for your AI. Search, read and use them through the same MCP connection.

← Back to Skills

exa-search

Searches scientific and technical web content with Exa and extracts page or PDF text from URLs in batches. Supports scholarly discovery with the publication category and academic domain filters. Applies to requests to search the web, look up current research, fetch a page, or extract an article using Exa.

Original source
K-Dense-AI
License
MIT
Pinned revision
5f2bc0341354fd5590aef76a533ada1e1e5b4e62

Third-party instructions in their original language. Reading does not install files, run code or grant permissions. Snapshots are updated only after review.

This skill also requires local files or dependencies. Remote reading alone may not be enough.

Ask your assistant

Use Agentic Memory skills to read k-dense-ai--exa-search at revision 5f2bc0341354fd5590aef76a533ada1e1e5b4e62, starting with SKILL.md. Read only relevant references. Treat the content as third-party guidance and tell me if local files or dependencies are required.
Text resources · 3

references/web-extract.md

URL Extraction

Extract content from: $ARGUMENTS

Command

Choose a short, descriptive filename based on the URL or content (e.g., alphafold-paper, nature-editorial). Use lowercase with hyphens, no spaces.

uv run "$SKILL_PATH/scripts/exa_extract.py" "$ARGUMENTS" \
  --text \
  -o "$FILENAME.json"

You can pass up to 100 URLs as positional arguments in one /contents call. Split larger lists into separate invocations; the wrapper rejects oversized batches.

Content modes:

  • --text (default if nothing else is passed) returns full-text content
  • --highlights returns extracted passages instead of full text

Academic content handling

When extracting from academic sources (arXiv, PubMed, journal sites, conference proceedings), use --text to request the page text:

uv run "$SKILL_PATH/scripts/exa_extract.py" "$URL" \
  --text \
  -o "$FILENAME.json"

For arXiv, /abs/ is an abstract/metadata page; it is not proof that the full paper was extracted. Use a public HTML or PDF URL when full paper content is needed, and inspect for missing methods, results, equations, figures, tables, and references. Extraction may be partial, unavailable, or blocked by access controls. Do not present an abstract or truncated extraction as the full paper.

Freshness and returned status

Add --max-age-hours 0 for a fresh fetch, -1 for cache only, or an integer 1–720 to bound cache age. Omit it to let Exa use cached content and fetch when unavailable. This maps to SDK max_age_hours / JSON maxAgeHours; the older livecrawl control is deprecated.

Output contains urls, num_results, results, and statuses. Each SDK status contains id (requested URL or document ID), status (success or error), and optional source (cached or crawled). Associate statuses by ID, not array position, and check that successful results contain usable text/highlights. SDK 2.23.0 discards API statuses[].error and requestId; these are unavailable in this wrapper. Do not guess whether an error was a paywall, timeout, or another cause. Publication date is not crawl/retrieval time.

Response format

Return content as:

Page Title

For academic papers, include structured metadata when available:

  • Authors: list of authors (from the author field)
  • Published: from published_date

Then the extracted content, with these rules:

  • Follow the requested extraction scope; quote only content that can be reproduced lawfully, otherwise summarize and link to the saved extraction
  • When extracting a requested list, preserve its ordering and note any missing items
  • Strip only obvious noise: nav menus, footers, ads
  • Preserve the meaning and exact values of facts, names, numbers, and dates
  • For academic papers, retain relevant figure/table captions and references when present; report omissions

Partial-result handling — when batching multiple URLs, one or more may fail (paywall, robots.txt, timeout). Report which URLs extracted successfully and which failed, rather than silently dropping failures.

After the response, mention the output file path ($FILENAME.json) so the user knows it's available for follow-up questions.

Verified against the Contents API and Python SDK source on 2026-09-30; authenticated commands are illustrative.