Quick start
There are two ways to start. Pick the one that matches what you want.
| You want… | Use |
|---|---|
| Your desktop agent (Claude Code, Claude Desktop, Cursor…) to read the web | Local binary over stdio: no server, no ports |
| A shared hub with an admin console, worker nodes and search | Docker hub: one container, no cloud account |
Option 1: Local binary for your MCP client
Section titled “Option 1: Local binary for your MCP client”go install github.com/bartrumb/edgefetch/v2/cmd/edgefetch@latestedgefetch initedgefetch doctorNo Go toolchain? Download a release archive for your OS from GitHub Releases instead.
Then register it with your MCP client:
edgefetch install-mcpThis writes a stdio server entry into Claude Desktop and Cursor. For other clients, see
Connect an MCP client.
Try a scrape from the terminal:
edgefetch scrape https://example.comOn the desktop, EdgeFetch uses the Chrome, Edge or Brave you already have installed to render JavaScript pages. It doesn’t download its own browser.
Option 2: Full hub in Docker
Section titled “Option 2: Full hub in Docker”git clone https://github.com/bartrumb/edgefetch.gitcd edgefetchdocker compose up --build -dOpen http://localhost:8787/admin. The local development token is printed in the repository
README. Change it before you expose the hub anywhere.
The all-in-one container includes the hub, a Go worker with headless Chromium, and SQLite
full-text search under /data. It needs no Redis and no Postgres.
Next steps
Section titled “Next steps”- Deploy a hub on Cloudflare Workers, a VPS or Home Assistant.
- Add worker nodes at home, without opening ports.
- Read the MCP tools reference.