Atomic Chat

What is Atomic Chat?

Atomic Chat is a free, open-source desktop and mobile app for running local LLMs and connecting MCP servers directly to your models. It acts as an MCP host: add a server once and its tools are available in every chat — local or cloud. It runs on macOS, Windows, Linux, iOS, and Android.

How to use Atomic Chat?

Download the app from https://atomic.chat. Open Settings → MCP Servers → Add, paste any MCP server config (name, transport: STDIO / HTTP / SSE, command or URL), then start chatting. The model automatically picks and calls tools. Works with any server from mcp.so and the official MCP registry.

Key features of Atomic Chat

- MCP host with in-app log viewer for tool calls
- Supports 1000+ local models from Hugging Face
- Connect cloud providers (OpenAI, Anthropic, Mistral, Groq, MiniMax) with your own key
- Provides an OpenAI-compatible API at http://localhost:1337/v1
- Privacy: local model data stays on your machine
- Integrates with Kilo Code, OpenCode, Goose, and other agents

Use cases of Atomic Chat

1. Connect Notion, GitHub, Google Drive, or web search to a local model
2. Run agents and coding tools on your own hardware via the local API
3. Mix local inference with cloud models without reconfiguring MCP servers

FAQ from Atomic Chat

Which MCP transports are supported?

STDIO, HTTP, and SSE.

Does my data go through Anthropic or OpenAI?

With a local model, tool calls go from your machine to the MCP server you connected — not through an AI provider.

Is Atomic Chat free?

Yes. It is open-source under Apache 2.0.

What platforms and models does it support?

It runs on macOS, Windows, Linux, iOS, and Android. Local models include Llama, Qwen, DeepSeek, Gemma, Mistral, and over 1,000 more from Hugging Face.