vishalsingh2972/namma-mcp

Namma MCP is a voice agent built for people in Bengaluru who want to talk to a computer program in natural, local language. Instead of typing commands, a person speaks into a microphone, and the agent listens, thinks, and replies out loud, mixing English and Kannada the way many people in the city actually talk. Under the hood, the project connects a few outside services together. LiveKit handles the live audio connection, carrying the sound back and forth over the internet in real time. Sarvam supplies the AI pieces: one model turns speech into text, a language model figures out what the person wants and produces a reply, and another model turns that reply back into spoken audio. There is also background noise filtering, so street sounds in a busy city do not confuse the agent. What makes this more than a simple chatbot is its use of the Model Context Protocol, or MCP. This lets the agent call small Python functions while it is still listening and thinking, so it can look something up, such as an order status, and speak back a real answer rather than a generic reply. The README walks through an example: someone asks about order number 502, and the agent checks a local function and answers with the delivery status, spoken partly in Kannada. To run it, a developer installs a handful of Python packages, sets up a LiveKit project, and adds API keys for LiveKit and Sarvam into a local settings file. Once configured, the agent can be tested from a computer's own microphone and speakers using a console command. The project is small: one main Python file holds the agent logic, alongside a requirements file and environment settings. It is aimed at developers who want a working example of a real-time, bilingual voice assistant with tool calling built in, rather than a polished consumer product.