Labsco
MCP SERVER

Skyvern

by Skyvern-AI

Hand your AI app a browser that fills forms, downloads files and researches the web — run locally or against Skyvern Cloud.

Browser Automation, Control & Screenshots
Summary
The remote endpoint is stateless — create a session first or browser calls fail.

This is the trap worth knowing before you debug it: on the remote `/mcp` endpoint, browser tools return `BrowserNotAvailable` unless you have called `skyvern_browser_session_create` and are passing the session id along. Running locally also means choosing and paying for your own model; cloud means an account and a key instead.

What it is

Skyvern's MCP server, the connector between MCP clients and Skyvern's browser automation. You can point it at a Skyvern server you run yourself with the model of your choice, or at Skyvern Cloud with an API key from the settings page.

What you get
  • Browser work driven from the conversation: filling out forms, downloading files, and researching information on the web
  • A setup wizard that writes the client configuration for you — `skyvern init` covers Cursor, Windsurf and Claude Desktop, and `skyvern setup opencode` handles OpenCode with API-key auth
  • A choice of backend: a local Skyvern server started with `skyvern run server`, or Skyvern Cloud
  • Explicit browser sessions on the remote endpoint — create one with `skyvern_browser_session_create` and pass its `browser_session_id` on every subsequent browser call
Requirements

Python 3.11 — the README states Skyvern only runs in that environment today. Install with `pip install skyvern`, run `skyvern init` to configure. `SKYVERN_BASE_URL` selects the backend: `https://api.skyvern.com` for cloud or `http://localhost:8000` when self-hosted, with `SKYVERN_API_KEY` alongside it. The server itself is launched as `python -m skyvern run mcp` over stdio. The package is `skyvern` 1.0.51.

Setup effort

One command plus a key — pip install skyvern, then supply credentials