Seven tools is all you need. A minimal AI agent that just works, using only seven tools. Comes with hands-free Turbo mode and Hot-reloading for self-modification.

Watch it in action:

Go Turbo: The standard pace is for chumps. Have it go full autonomous by disabling ask.

# @mcp.tool()
async def ask(question: str) -> str:

Devcontainer (.devcontainer) included and useable, otherwise proceed with manual install.

1. Provide credentials in servers/.env (used for browsing LLM) and clients/.env (used by agent) file accordingly.

2. Install UV:

# OSX/Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# Windows
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

3. Start the MCP server:

# --reload enables Hot-reloading
cd servers && PYTHONPATH=. uv run app.py --reload

# For Browser Use (on linux)
# sudo apt-get update && sudo apt-get install -y  libglib2.0-0t64 libnss3 libnspr4 libdbus-1-3 libatk1.0-0t64 libatk-bridge2.0-0t64 libcups2t64 libxkbcommon0 libatspi2.0-0t64 libxcomposite1 libxdamage1 libxfixes3 libxrandr2 libgbm1 libpango-1.0-0 libcairo2 libasound2t64
# cd servers && uv run sync && uv run patchright install chromium

4. Start the MCP client (self-executable UV script):

# For native Gemini client
./clients/gemini_client.py http://localhost:8081/sse # --debug

# For OpenAI-compatible clients (including Anthropic)
./clients/openai_client.py http://localhost:8081/sse # --debug

• Delete content_history.json to clear message history.

• OpenAI compatible APIs (including Anthropic)
• System prompt guidelines with single file project templates

• This is a customisable sharp tool for now. Guardrails will only be implemented over time.

For enquiries, feel free to reach out to [email protected].