Tool design guide / 2026
Designing AI Agent Tool Calls for Search and Retrieval in 2026
An agent performs better when tools reflect intent instead of exposing a grab bag of transport details. Clear names, validated inputs, predictable shapes, and explicit scope make behavior easier to evaluate.
Quick answer
Make each tool answer one retrieval question.
Use search for “find sources,” extraction for “read this URL,” news for “what is recent,” and finance for a current market snapshot.
curl --get 'https://extractor.sh/api/extract' \
--data-urlencode 'url=https://example.com/' \
--data-urlencode 'format=json'Available data
What you can extract
- Purpose-built tool inputs
- Markdown and JSON representations
- Shared cache and rate-limit behavior
- Source-linked output
AI workflows
Where normalized data helps
- MCP client integrations
- Function-calling chatbots
- Agent evaluation harnesses
- Developer tooling assistants
AI-ready output
Markdown for models. JSON for systems.
Raw HTML consumes tokens on navigation, scripts, styling, and interface labels. Clean Markdown keeps the readable hierarchy for LLM prompts and RAG chunks. Normalized JSON is better when your application needs an explicit semantic type, source, author, publication date, media, attributes, and collection items.
Always retain the canonical URL from the response. AI-generated summaries should remain traceable to the public source, especially when the underlying page can change.
Boundaries
Public data only
- Tools cannot resolve ambiguous intent without a clear assumption.
- Do not expose credentials through retrieval tools.
- No tool promises restricted-source access.
extractor.sh does not bypass CAPTCHAs, login walls, paywalls, access controls, or regional restrictions. Review the source’s terms and applicable law before collecting or reusing data.