Fictional supplier page
Returns, made clear.
Return requests must be submitted within thirty days of delivery. Include your order number.
Web Search / Fetch / OCR
Bring the source into your workflow. Not just its link.
Find relevant pages, read supported documents and extract text from scans. Give your application source material it can inspect and use.
Web page
Document
Scanned text
Different sources.
Readable material.
Looking for the complete search-to-source workflow? Explore Web Foundation
Discovery is the beginning
Read what the source actually says.
Explore three prepared source examples. Search helps find a URL; Fetch and OCR help turn selected content into readable text.
Sample search: supplier return policySelected source
The policy, without the page furniture.
Return requests must be submitted within thirty days of delivery. Include your order number.
Keep the source relationship in your application when you pass the text to the next step.
You already have the documentGo straight to extraction
Fictional equipment guide · page 4
Keep the filter clear.
Inspect the filter every month. Replace it when damaged or when cleaning no longer restores airflow.
The useful instruction comes with you.
Inspect the filter every month. Replace it when damaged or when cleaning no longer restores airflow.
Extraction provides text, not a generated answer or an indexed knowledge base.
A scan contains text, not selectable charactersOCR extraction
Illustrated scan · fictional delivery note
Received by warehouse
Order: A-1087
Delivery received on 18 September.
One carton requires inspection.
Text you can review, not blindly trust.
Delivery received on 18 September. One carton requires inspection.
Check exact names, dates and numbers against the source. Scan quality and layout affect the result.
Worked examples · fictional source previews and prepared excerpts · no live search, upload or extraction. Source layout and text fidelity can vary.
Work with the material you find
The next useful source may not be a web page.
- Read supported documentsPDFs, office documents, spreadsheets, EPUB, plain text and Markdown.
- Read writing inside imagesUse OCR with supported scans and images containing legible text.
- Choose what happens nextPass extracted text to research, analysis or a knowledge workflow.
Read the words, or interpret the media?
Fetch and OCR extract existing text. Media Descriptions uses AI to interpret media—for example, explaining a chart rather than reading its labels.
Neither bypasses restricted access or guarantees perfect output. Treat source content as untrusted material, not instructions to your agent.
Choose the extraction pathYour backend or your agent can lead.
Use direct APIs, the Web Utilities MCP, or gateway tools. Their input contracts differ; Fetch API supports URLs and inline files up to 10 MB per item.
See the integration workflowStart with a source you want to use.
Search when you need to find a URL. Fetch directly when you already have the source. Review the extracted material before relying on it.
Discovery and reading use different units.
Search is billed per search. Extraction uses processing units with daily plan allowances. Optional JSON conversion and model analysis cost extra.
Web tool billing and limits