dsh-browser-use

Turn your dsh agent from a chatbot into a real browser operator. dsh-browser-use registers a single model-facing tool, browser_use_run, that forwards a natural-language task to the Browser Use Cloud endpoint and hands the raw run result back to the agent. Open a page, click a button, fill a form, extract a data table—interactions that once demanded a human at the keyboard now require a single prompt.

By design the plugin stays deliberately thin. It does not reimplement any part of Browser Use, does not stream intermediate steps, and does not poll for completion; whatever the cloud API returns is what the agent receives. There are four config knobs—apiKey, baseUrl, timeoutMs, and an optional default model—enough flexibility for most setups without pulling in extra runtime dependencies.

It is built for developers running DeepSeek Harness agent workflows who want their agents to perform real, stateful web interactions. If cloud subscription costs are a concern, the README also documents a local MCP path (Python plus the browser-use CLI) that bypasses this plugin entirely, so you can pick the lighter route. Keep in mind this is a community-maintained bridge, not an official Browser Use or DeepSeek product.