Ask a page a question. jurl prints only the part that answers it, word for word, with a link to where it says it. When the answer is pages away, it goes looking.
Open source · MIT or Apache-2.0 · GitHub
$ jurl --precise --follow 15 \ -q "What is the boiling point of mercury?" \ en.wikipedia.org/wiki/Tennis 629.88 K (356.73 °C, 674.11 °F)
Every line it prints is on the page as you see it. No summary, no invented URL.
No answer, exit code 1. A script gets an empty cell, not a wrong one.
Most pages, end to end. About $4 per 10,000 pages with your own keys.
--preciseOne value, for a spreadsheet cell or a script.
$ jurl --precise -q "what is the primary rate limit for authenticated users?" docs.github.com/en/rest/using-the-rest-api/rate-limits-for-the-rest-api 5,000 requests per hour <https://docs.github.com/en/rest/…#:~:text=rate%20limit%20of-,5%2C000%20requests%20per%20hour>
--followGive it the bare domain. It finds the page.
$ jurl --follow --precise -q "What is the monthly price of the cheapest paid plan?" linear.app $10 per user/month <https://linear.app/pricing#:~:text=%2410%20per%20user%2Fmonth>
--findLooks at the pixels, captions or not.
$ jurl --find "a cathedral" -n 3 en.wikipedia.org/wiki/Cologne https://thumb.wikimedia.org/…/2/27/Kdom.jpg/250px-Kdom.jpg https://thumb.wikimedia.org/…/K%C3%B6ln_um_1890.jpg/500px-K%C3%B6ln_um_1890.jpg https://thumb.wikimedia.org/…/Kranh%C3%A4user_Cologne%2C_April_2018_-01.jpg
--linksOnly the links that match what you asked.
$ jurl --links -q "essays about startups" -n 4 paulgraham.com/articles.html https://paulgraham.com/start.html https://paulgraham.com/startupmistakes.html https://paulgraham.com/startupideas.html https://paulgraham.com/13sentences.html
Also --code for just the code, -q "…" for one answer, --pages to read a list to its end, and plain jurl <url> for the gist. Pages built with JavaScript just work. Every flag →
Ten documentation pages, one question each, read by jurl and by the tools agents use today.
| Measured | jurl | WebFetch | Exa | Tavily |
|---|---|---|---|---|
| Exact answer | 10/10 | 10/10 | 8/10 | 5/10 |
| Code lines not on the page | 0 of 60 | 33 of 63 | 0 of 59 | 8 of 36 |
| Cost per 10,000 pages | $4 | ~$150 | $10 | $16 |
| Finds an image by what it shows | yes | no | no | no |
Measured on 2026-10-07. WebFetch is Claude Code's. The tasks, the scripts and every answer →
New in 0.2.0 · jurl cloud
Sign in once and jurl reads with jurl cloud: no API keys, the same flags and the same output (photos still read with your own keys). Any answer can become a sheet that jurl reads again on a schedule.
$ jurl login To sign in to jurl cloud, open https://cloud.jurl.dev/device and enter this code: KQ7M-PX4T Signed in as you@acme.com to Acme. Console: https://cloud.jurl.dev $ jurl --follow --precise -q "What is the monthly price of the cheapest paid plan?" linear.app $10 per user/month jurl: keep it up to date in a sheet: https://cloud.jurl.dev/w/…
| Site | Cheapest paid plan | Read |
|---|---|---|
| linear.app | $10 per user/month | 09:12 |
| + add sites | ||
| + read it again every day | ||
$ brew install rayoplateado/tap/jurl$ curl -LsSf https://github.com/rayoplateado/jurl/releases/latest/download/jurl-installer.sh | shpowershell -ExecutionPolicy Bypass -c "irm https://github.com/rayoplateado/jurl/releases/latest/download/jurl-installer.ps1 | iex"$ claude mcp add jurl -- jurl mcpA free account with a monthly allowance of page reads. Nothing to set up.
jurl loginA TypeSafe key, checked and saved on the first run. You pay per page.
jurl initWhat leaves your machineWith your keys, the page's text goes to TypeSafe, and with --vision or --find its photos go to Cloudflare. With jurl cloud, jurl's servers read the page.
What it can't readPages behind a login: jurl never sends your cookies.
Lightpanda, which jurl downloads for JavaScript pages, is a separate program under AGPL-3.0.