Home

fetchwright reads public web pages that refuse scripts, through Gemini's URL context tool. A Reddit thread, for one, answers a plain HTTP client with a block page. fetchwright hands the URL to Gemini, which reads the page from Google's own index or with a live fetch and gives it back as JSON. A small command-line tool then checks that Google retrieved the page, drops what a note should not keep, and prints the result.

It is an agent plugin (Claude Code, and any agent that reads Agent Skills) with two skills and one standard-library Python script. Version 0.2.0, released 2026-10-16.

It is not a crawler or a search tool. It reads one URL at a time, never searches, and never uses Google Search. It cannot read pages behind a login or a paywall, YouTube, Google Docs, or private and local addresses.

The smallest example

python fetchwright.py read https://www.reddit.com/r/examplesub/comments/1abc234/

prints:

# I made a small upscaler node
url: https://www.reddit.com/r/examplesub/comments/1abc234/
model: gemini-3.8-flash  mode: thread  style: verbatim
retrieval: success https://www.reddit.com/r/examplesub/comments/1abc234/a_small_upscaler_node/; error https://www.reddit.com/r/examplesub/comments/1abc234/
tokens: page 4210, output 410, total 4810

It upscales video in tiles so it fits in 8 GB.

Link in the comments.

## Comments (3 kept; dropped 1 placeholders, 1 lone usernames or words, 0 empty)
1. Works on my 3060, about 2 fps at 1080p.
2. [OP] Thanks for testing. Tiles of 512 are the default.
3. Does it keep the audio track?

The thread and its text are sample content. Every output in this wiki comes from fetchwright 0.1.0 run against a stand-in for Google's API that answers in the shape the real API used on 2026-10-15, so the pages show exact output without spending anyone's quota. The URL above is made up. A real read looks the same, with real text and timings.

Pages

• Getting started: install the plugin, get and store a key, the first read.
• Commands: every option of check and read, which stream each message goes to, and the exit codes.
• How pages are read: threads and pages, the retrieval check, what gets dropped, models and fallback, the JSON record.
• Terms and privacy: what Google's terms say, what the free tier means for what you send, and what is untested.
• Recipes: save a thread as JSON, call fetchwright from another program, read several URLs, cut costs.
• FAQ: questions you are likely to have.
• Versions and upgrading: the releases.
• Development: tests, the size budget, how to change the CLI.