# jurl CLI: curl that reads the page for you 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. ``` $ 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) ``` A recorded run: 13 pages read, 8 dropped, 3.8 s. The path it took: Tennis → Kevlar → Chemical formula → Periodic table → Mercury (element). - Install: `brew install rayoplateado/tap/jurl` - New in 0.2.0: `jurl login` reads with jurl cloud, no API keys: https://jurl.dev/cli/#cloud - Code: https://github.com/rayoplateado/jurl (MIT or Apache-2.0) - All flags: https://github.com/rayoplateado/jurl#reference - Changelog: https://github.com/rayoplateado/jurl/blob/main/CHANGELOG.md - Blog: https://jurl.dev/blog/ (every post also in markdown: add `.md`) ## Why - **Picks, never writes.** Every line it prints is on the page as you see it. No summary, no invented URL. - **Says when it isn't there.** No answer, exit code 1. A script gets an empty cell, not a wrong one. - **Under a second a page.** Most pages, end to end. About $4 per 10,000 pages with your own keys. ## One command, a flag for each job ### The exact answer: `--precise` One value, for a spreadsheet cell or a script. The link highlights the sentence. ``` $ 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 All of these requests count towards your personal rate limit of 5,000 requests per hour. … ``` ### Find it on the site: `--follow` Give it the bare domain. It finds the page. Same site only, `robots.txt` respected, 5 pages unless you say more (`--follow 15`). ``` $ jurl --follow --precise -q "What is the monthly price of the cheapest paid plan?" linear.app $10 per user/month ``` 2.4 s. Link: https://linear.app/pricing#:~:text=%2410%20per%20user%2Fmonth,-Billed%20yearly%20All ### A photo, by what it shows: `--find` Looks at the pixels, captions or not. Wikipedia's Cologne article has 70 images: ``` $ jurl --find "a cathedral" -n 3 en.wikipedia.org/wiki/Cologne https://thumb.wikimedia.org/wikipedia/commons/thumb/2/27/Kdom.jpg/250px-Kdom.jpg?… https://thumb.wikimedia.org/wikipedia/commons/thumb/0/0a/K%C3%B6ln_um_1890.jpg/500px-K%C3%B6ln_um_1890.jpg?… https://thumb.wikimedia.org/wikipedia/commons/thumb/5/5f/Kranh%C3%A4user_Cologne%2C_April_2018_-01.jpg/1280px-Kranh%C3%A4user_Cologne%2C_April_2018_-01.jpg?… ``` 2.9 s. The third match has no alt text and no caption: only its pixels say there's a cathedral in it. When nothing on the page matches, jurl says so and exits with code 1. ### Links worth following: `--links` ``` $ 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 ``` 0.8 s. Safe to pipe into `xargs curl`. Also: `--code` for just the code, `-q "…"` for one answer, `--pages` to read a list to its end, and plain `jurl ` for the gist. Pages built with JavaScript just work. ## Never rewrites the page Ten documentation pages, one question each, read by jurl and by the tools agents use today. | | jurl | Claude Code 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. The tasks, the scripts and every answer: https://github.com/rayoplateado/jurl/tree/main/bench ## New in 0.2.0: an answer today, a sheet that stays true Sign in once and jurl reads with jurl cloud: no API keys, same flags, same output. At a terminal, an answer ends with a link that turns it into a sheet in jurl cloud, which reads it 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/… ``` `jurl status` shows the account and the page reads used; `jurl logout` signs out. jurl cloud doesn't take `--image`, `--vision` or `--find`: those read with your own keys. ## Install ``` brew install rayoplateado/tap/jurl # macOS, Linux curl -LsSf https://github.com/rayoplateado/jurl/releases/latest/download/jurl-installer.sh | sh # no Homebrew ``` Windows: `powershell -ExecutionPolicy Bypass -c "irm https://github.com/rayoplateado/jurl/releases/latest/download/jurl-installer.ps1 | iex"` · From source: `cargo install --git https://github.com/rayoplateado/jurl` Then pick how it reads: - **jurl cloud**: `jurl login`. A free account with a monthly allowance of page reads. Nothing to set up. - **Your own keys**: `jurl init`, or just the first run. A TypeSafe key (https://console.typesafe.ai), checked and saved. You pay per page. For `--vision` and `--find`, add a Cloudflare Workers AI token. For your agent: `claude mcp add jurl -- jurl mcp` gives Claude Code, Claude Desktop or Cursor jurl as a tool. - **What leaves your machine:** with 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 read:** pages behind a login. jurl never sends your cookies. --- Made in Murcia by Ray García (@rayoplateado on X). MIT or Apache-2.0. Lightpanda, which jurl downloads for JavaScript pages, is a separate program under AGPL-3.0.