JARVIS Chat Agent

A Python-based chat runtime for JARVIS. It handles natural language processing, tool orchestration (ReAct), and routes all hardware/system commands to the Rust MCP server.

Overview

• Runtime: Python 3.10+ (managed via uv)
• Backend: Communicates with jarvis-skills/rust-mcp-server via JSON-RPC.
• Tools: Dynamic loading from MCP tools/list with smart local caching.

Setup

1. Install dependencies:

   cd jarvis-chat
   uv sync

2. Configure environment: Create a .env file from the example and add your keys:

   cp .env.example .env

Running the Agent

1. Start the Rust MCP Server

(From the jarvis-skills directory)

cd rust-mcp-server
cargo run --release

2. Launch the Chat Agent

cd jarvis-chat
python main.py

• Interactive mode: Just run python main.py and start typing.
• Server mode: Run python main.py --server --port 8000 to expose the HTTP API.

Verification

Check if everything is connected correctly:

1. Health Check: curl http://127.0.0.1:5050/health
2. Tool Discovery: curl http://127.0.0.1:5050/tools
3. Smoke Test: python main.py "get system info"

Copilot CLI / Gemini CLI MCP config

This repo now includes .mcp.json with STDIO server launch commands for local CLI testing.

Prerequisites:

cd ../spotify-mcp-server
npm install
npm run build

Then run your CLI from jarvis-chat so relative paths in .mcp.json resolve correctly.

Notes

• Caching: LLM responses are cached for 3 minutes to save on API costs and latency.
• Routing: Complex hardware intents (Spotify, Volume, WiFi) are routed directly to Rust for native execution.
• Requirements: Ensure the Rust MCP server is running before starting the Python agent.