Skip to content
AnyCapAnyCap Docs

Research the Web with Search and Crawl

1. Narrow the search space

Section titled “1. Narrow the search space”

Start with structured results and skip full-page crawling while you classify sources:

Terminal window
anycap search \
  --query "MCP protocol release notes" \
  --no-crawl \
  --max-results 10 \
  | jq -r '.data.results[] | "\(.title) — \(.url)"'

Add --include <domain> for authoritative sources or --time-range month when recency matters.

2. Read only the useful pages

Section titled “2. Read only the useful pages”
Terminal window
anycap crawl https://example.com/relevant-page \
  | jq -r '.data.markdown' > source.md

Crawl known HTML pages; use a download tool for PDFs, images, archives, or other binary resources.

3. Use grounded search for synthesis

Section titled “3. Use grounded search for synthesis”

When the task needs an answer across multiple current sources:

Terminal window
anycap search \
  --prompt "What changed in the latest MCP specification, and what affects local stdio servers?" \
  > grounded-answer.json

Inspect both the answer and its source mapping:

Terminal window
jq -r '.data.content' grounded-answer.json
jq -r '.data.search_metadata.sources[] | "\(.index). \(.title): \(.uri)"' grounded-answer.json

4. Preserve evidence

Section titled “4. Preserve evidence”

Record the query, retrieval date, URLs, relevant excerpts, and diagnostic request_id. Treat synthesized output as an aid to analysis, not a substitute for checking high-impact claims against primary sources.