Browser use agent with Chrome Dev Tools MCP and Langchain
Small LangGraph/LangChain agent that drives the browser through the Chrome DevTools MCP client. Uses DeepSeek for the LLM, Ollama embeddings for memory, and stores state in SQLite/Chroma under ./mem.
What you need
- Python 3.13+
uvCLI (pip install uv)- Chrome installed (for the MCP tool)
- Ollama running locally with the
nomic-embed-textmodel, or change the embedding provider incu/utils/__init__.py
Setup and run
1) Copy .env_example to .env and set at least DEEPSEEK_API_KEY. LANGSMITH_API_KEY is optional; OPENROUTER_API_KEY is only needed if you switch providers. 2) From the repo root run ./run.sh (Linux/Mac) or ./run.ps1 (Windows). The script will create .venv with uv venv --python=3.13 if missing, install requirements.txt, and start python -m cu.agent. 3) Type your requests in the REPL; type exit to quit.
Notes
- Set
STORAGE_LOCATIONin.envif you want memory/checkpoints in a different folder. - If
uvis missing, install it first (pip install uv) before running the scripts.
