Turn web pages into research-ready Markdown
Three official pages, three readable results and $0.009 in data fees. Keep the source and its meaning when you turn a reading list into notes.
Research ·
Give the agent a document it can work with
A page URL is a useful starting point, but research depends on reading the body, not just the search snippet. Converting a public page to Markdown makes headings, paragraphs and links easier to inspect, quote carefully and carry into a sourced note.
We passed three Linear URLs to contextdev.scrape_markdown: the pricing page, the Series C announcement and a later post about growth and employee liquidity. Each returned a successful Markdown result. The pages retained the distinctions that mattered to the research, including annual-billing language, announcement dates and the description of the tender offer.
The extracted text still included some site navigation. A successful conversion is not the same as a perfectly cleaned document. Check the content you plan to use before handing it to the next step.
Save the source beside the text
Keep the URL, title, returned Markdown, tool run ID and extraction status together. Those fields let a reader find the original and let you diagnose whether a later mistake came from retrieval or interpretation. A collection timestamp belongs in your working archive so you can judge whether a fresh read is needed.
Keep the unedited response as well as the cleaned note. Remove repeated menus when preparing a research document, but preserve qualifiers and links in the body. Removing the words that explain an annual price or a secondary share sale would make the note shorter and less accurate.
This example covers public HTML pages. It does not demonstrate PDF parsing, OCR, private pages or a whole-site crawl. Choose a tool for the actual source type before extending the same workflow to a mixed reading list.
The three-page result
Each page cost $0.003, for $0.009 in AR data fees. We requested a fresh fetch with maxAgeMs set to zero. For stable material, the tool also accepts a cache-age allowance; check its current contract before deciding how often you need to retrieve the same source.
These page reads also supported the company-brief and pricing examples. Reusing a saved result does not create another provider call. Do not add all the articles’ example subtotals together as if each shared page had been fetched again.
| Source | Result | Actual charge |
|---|---|---|
| Pricing page | Markdown returned | $0.003 |
| Series C announcement | Markdown returned | $0.003 |
| Growth and tender-offer post | Markdown returned | $0.003 |
| Total | 3 pages | $0.009 |
Build a small reading workflow
Start with a few known URLs and inspect one full result before processing the rest. Confirm that the body contains the material you came for, rather than a login screen, an error or only navigation. Keep failures visible and separate them from usable documents.
For a research brief, preserve the original documents and write a separate synthesis with source links next to factual claims. For a retrieval collection, keep the source metadata with each chunk. Either way, retaining the original page text makes it possible to revisit a conclusion without paying to fetch the same evidence again.
Using AgentRouter, convert https://linear.app/pricing, https://linear.app/now/building-our-way and https://linear.app/now/sharing-growth-with-the-people-building-linear to Markdown. Save the full responses with source URLs and run IDs. Check that each contains the page body. Make a separate short research note that preserves billing conditions and distinguishes a funding round from a tender offer. Report failures and actual charges. Keep AR data fees under $0.02.