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-search.md

Web Search

Search the web for: $ARGUMENTS

Command

Choose a short, descriptive filename based on the query (e.g., ai-chip-news, crispr-off-target). Use lowercase with hyphens, no spaces.

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

$SKILL_PATH is the path to this skill directory. The -o flag saves the full results to a JSON file so follow-up questions can reuse them without re-querying.

Search type selection — --type controls retrieval mode:

| Mode | When to use | |---|---| | auto (default) | Exa's general-purpose search. Use this unless you have a reason not to. | | fast | Low-latency interactive search. | | instant | Minimum response time, trading some depth for speed. | | deep-lite | Lightweight research with lower latency than deep. | | deep | Multi-step research for complex queries. | | deep-reasoning | Higher-effort reasoning for complex analysis. |

Content modes — add any combination:

  • --text returns full-text content per result
  • --highlights returns the most relevant passages (good signal-to-noise, lower token cost than full text)

Default to --highlights for broad searches (less text to process). Add --text only when you need to quote or extract in detail.

Filtering options — Exa supports rich filtering via the SDK:

  • --start-published-date 2026-01-01T00:00:00Z / --end-published-date 2026-10-01T00:00:00Z for ISO 8601 publication-date bounds
  • --include-domains domain1.com,domain2.com to restrict to an allowlist
  • --exclude-domains spam.com,low-quality.com to drop a blocklist
  • --category publication to bias toward scholarly content (also: company, news, personal site, financial report, people; other strings are category hints)
  • --user-location US for locale-specific results

--num-results accepts 1–100 (individual modes may impose lower limits). There is no documented offset/cursor pagination; reformulate a query or partition by supported filters when more coverage is needed. Neither method establishes an exhaustive literature search. company and people reject publication-date and exclude-domain filters; the wrapper catches these combinations before calling. Use --include-domains github.com instead of the deprecated github category. Domain filters also support paths and wildcard subdomains. No content flag means metadata only, despite the SDK's default text retrieval.

Academic source strategy

For scientific or technical queries, Exa has two strong levers:

1. Use --category publication

uv run "$SKILL_PATH/scripts/exa_search.py" "$ARGUMENTS" \
  --category publication \
  --text --highlights \
  -o "$FILENAME-academic.json"

This selects the current scholarly-publication category (papers, preprints, and journal articles). Verify peer review, publication metadata, and relevance on the source; the category does not establish evidence quality.

2. Restrict to scholarly domains

For stricter academic filtering, combine the category with an explicit domain allowlist:

uv run "$SKILL_PATH/scripts/exa_search.py" "$ARGUMENTS" \
  --category publication \
  --include-domains "arxiv.org,biorxiv.org,medrxiv.org,pubmed.ncbi.nlm.nih.gov,nature.com,science.org" \
  --text --highlights \
  -o "$FILENAME-academic.json"

Two-pass pattern for comprehensive coverage

Run both an academic-focused search and an unrestricted one, then merge with academic sources first:

  1. Academic pass: --category publication with the scholarly domain allowlist above.
  2. General pass: the standard command without --category or --include-domains, to catch relevant non-academic sources (news coverage, lab blogs, institutional pages).

Merge results, leading with academic sources. If the query is clearly non-scientific, skip the academic pass.

When to use the two-search pattern: Any query involving scientific claims, medical information, research findings, technical mechanisms, statistical data, or anything where primary literature would be more reliable than secondary reporting.

Parsing results

Parse the JSON output. Each result includes:

  • title, url, published_date, author
  • text (if --text) and highlights (if --highlights)
  • Legacy compatibility fields score, highlight_scores, and top-level autoprompt_string may be null/empty; never require them or infer evidence quality from a retrieval score. Exa's changelog says highlightScores was removed in May 2026 although the generated reference still lists it.

Missing author/date/text metadata is normal. The wrapper exports a selected subset of the SDK response, not raw API JSON or structured publication entities.

Snippet fallback — any combination of content fields may be present. Cascade through them: prefer highlights (tight, pre-selected passages), fall back to a truncated slice of text. Never assume exactly one is present.

Response format

CRITICAL: Every claim must have an inline citation. Use markdown links pulling only from the JSON output. Never invent or guess URLs.

For academic sources, use author-year citation style where metadata is available:

Synthesize a response that:

  • Leads with findings from peer-reviewed or preprint sources when available
  • Clearly distinguishes between claims backed by primary research vs. secondary reporting
  • Includes specific facts, names, numbers, dates
  • Cites every fact inline — do not leave any claim uncited
  • Organizes by theme if multiple topics
  • Notes the evidence quality (e.g., "a randomized controlled trial found..." vs. "a blog post reports...")

End with a Sources section listing every URL referenced, grouped by type:

Sources:

Academic / Peer-reviewed:
- [Smith et al., 2025 — Title of Paper](https://doi.org/...) (Nature, 2025)
- [Jones & Lee, 2024 — Title of Paper](https://arxiv.org/...) (arXiv preprint)

Other:
- [Source Title](https://example.com/article) (Feb 2026)

This Sources section is mandatory. Do not omit it. If no academic sources were found, report the search coverage and filters used; do not infer that the topic has not been studied.

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

Verified against the Search API and changelog on 2026-09-30; authenticated commands are illustrative.