I built Reflex, a browser automation MCP server, and wanted to share it here since Cursor speaks MCP.
The problem it targets: when an agent drives a browser, every look at a page returns the whole accessibility tree, and the model has to read all of it before it can act. On a heavy page that is tens of thousands of tokens for one look, paid on every turn.
Reflex sends one compact view of the page, then only what changed, and batches a whole task into one guarded call. In agent driven tests across 5 live tasks it finished about 3.5x faster end to end than Playwright MCP, usually 2 to 3 calls instead of a dozen. You can also give any call a session name and it opens a separate isolated browser, so several agents can drive in parallel instead of sharing one tab.
Honest caveat, and it matters most for Cursor users: if you are comfortable in the terminal, the Playwright CLI and Vercel’s agent browser are great and free, and on a flow you already planned the CLI sends even fewer tokens. Reflex earns its place when you want the agent itself to drive the browser through MCP without dropping to a shell, or when you want parallel isolated browsers.
Setup is the standard stdio MCP config, no build step: npx -y reflex-browser@latest mcp
Free tier is 30 credits, no card: https://reflexmcp.com
Would love feedback, especially where it breaks for you.