Browserbase

Browserbase’s MCP server hands an agent a real, remote browser instead of one running on the local machine — useful for environments where installing Playwright and its browser binaries isn’t practical, or where automation needs to run somewhere other than the developer’s own laptop (a CI runner, a serverless function, a sandboxed agent host). Session management, navigation, page interaction, and content extraction are exposed as six tools running against Browserbase’s cloud infrastructure rather than a local process.

Stagehand — Browserbase’s higher-level automation layer — sits underneath, translating natural-language page actions (“click the login button,” “extract the pricing table”) into concrete browser operations. That layer defaults to Gemini as its model but accepts a --modelApiKey flag to swap in another one, decoupling the choice of automation-instruction model from whichever agent is driving the MCP session itself.

Two required variables, BROWSERBASE_API_KEY and BROWSERBASE_PROJECT_ID, are the only setup cost. It’s distributed via npx or a Docker build, and supports both clients with native streamable-HTTP transport and older ones via the mcp-remote wrapper — a reasonable pick specifically when local browser automation isn’t an option, with Playwright MCP remaining the more common default when it is.