crawlio-browser-guide
v0.7.0
Published
Companion CLI and read-only MCP server for Crawlio Browser Guide: install the native helper with one command, and let Claude Code or Codex see the browser page the user chose to share, with no way to act on it.
Maintainers
Readme
crawlio-browser-guide
The companion CLI and local MCP server for Crawlio Browser Guide. It installs the native helper with one command and lets coding agents on your machine (Claude Code, Codex, or any MCP client) see the browser page you chose to share, without giving them any way to act on it.
Install the helper
npx crawlio-browser-guide init # stage the helper + register it with Chrome (macOS, Windows, Linux)
npx crawlio-browser-guide doctor # check every link, including a live host ping
npx crawlio-browser-guide uninstall # remove the registration (credentials stay)On macOS, init stages the universal (arm64 + x86_64) Swift helper shipped inside this package, signed with a Developer ID, and writes the Chrome native-messaging manifests byte-identically to the Swift installer. On Windows and Linux it stages the package's own Node helper, which passes the exact protocol conformance suite the Swift host does. It is idempotent; rerun it anytime.
How it works
While the Eyes toggle in the Browser Guide side panel is on, the extension publishes the sanitized snapshot of the shared tab (origin, title, bounded accessibility outline, never input values, passwords, or screenshots) to ~/.config/browser-guide/eyes.json (user-only permissions) through the native helper. This server only reads that file:
- Toggle off (the default) deletes the file: an absent file is the fail-closed state, and this server reports "eyes are off".
- Every response carries
captured_atso the agent knows how stale the view is. - The server has no browser connection, no network access, and exactly one read-only tool.
Register
claude mcp add browser-guide -- npx -y crawlio-browser-guide mcpOr in any MCP client config: command npx, arguments ["-y", "crawlio-browser-guide", "mcp"].
Tool
get_current_page: returns the origin, title, capture time (with a staleness warning past two minutes), and the sanitized page evidence, wrapped in explicit untrusted-content markers.
Privacy
Everything stays on your machine. See the Browser Guide privacy policy.
