Quickstart

Install

npm i -g webctx
webctx --help

Try the three commands

Search the web:

webctx search "agent web research"

Turn a URL into useful text:

webctx read-link https://github.com/amxv/webctx

Discover the pages on a site:

webctx map-site https://docs.firecrawl.dev

All three return plain text or markdown that can go straight into an agent, a file, a pipe, or your clipboard.

The part worth trying first

read-link understands many GitHub URLs instead of scraping the whole GitHub page.

# Just these source lines
webctx read-link 'https://github.com/amxv/webctx/blob/main/README.md#L1-L20'

# Just this Issue and its conversation
webctx read-link https://github.com/amxv/webctx/issues/6

# A PR as a review conversation
webctx read-link https://github.com/amxv/webctx/pull/15

# A single inline review thread
webctx read-link 'https://github.com/cli/cli/pull/13250#discussion_r3118513169'

The URL already carries your intent. webctx tries to preserve that intent and return only the useful context.

Add credentials

Create .env.local in the directory where you normally run webctx:

BRAVE_API_KEY=...
TAVILY_API_KEY=...
EXA_API_KEY=...
FIRECRAWL_API_KEY=...
GH_TOKEN=...

You do not need every key for every command:

What you want to do	Key
Normal search	BRAVE_API_KEY, TAVILY_API_KEY, EXA_API_KEY
Map sites or crawl normal pages	FIRECRAWL_API_KEY
More GitHub capacity / auth-only GitHub views	GH_TOKEN or GITHUB_TOKEN

Many public GitHub reads work without a GitHub token. Add one when you want higher limits, blame, Discussions, some Actions logs, private resources your token can access, or richer review-thread state.

A simple agent loop

webctx search "next.js caching changes"
webctx read-link https://nextjs.org/docs/app/building-your-application/caching
webctx read-link https://github.com/vercel/next.js/issues/<issue-number>

Search finds candidates. read-link turns the useful candidates into focused context. Your agent gets evidence instead of browser UI.