The Uplink MCP server (Documentation Index
Fetch the complete documentation index at: https://docs.uplink.build/llms.txt
Use this file to discover all available pages before exploring further.
@uplink-code/mcp) exposes Uplink’s browser automation surface to any AI client that speaks the Model Context Protocol. With it, an LLM running in Claude Code, Cursor, or Windsurf can pair a real mobile device, navigate pages, run JavaScript, and inspect the network traffic that the page produces — useful for reverse-engineering how a consumer website actually works.
When to use it
Reverse-engineering
Have an AI assistant drive a logged-in mobile session and observe the real API calls a site makes.
Investigative automation
Ad-hoc “show me how X works” workflows without writing a script first.
AI agent prototyping
Test an automation idea interactively before committing it to a JavaScript SDK script.
Live debugging
Ask the model to drive a flow and surface the network requests or console logs that broke.
What it isn’t
The MCP server is not a wrapper around an official OAuth API. Uplink drives a real device’s browser; the AI inspects whatever traffic the site itself produces. There’s noclient_id / client_secret step for the site you’re investigating — the user simply logs in interactively on their paired device.
How it works
Next steps
Install + sign in
Get the MCP server running in Claude Code or your preferred AI client.
Tool reference
Full catalog of
uplink_* tools the AI can call.