MCP Connectors
for Local AI
Atomic Chat uses MCP (Model Context Protocol) to connect Notion, Google Drive, GitHub, web search, or any other tool directly to a local model on your hardware. Ask about your documents, run live web searches — your prompts never pass through an AI provider’s servers.
Connect any tool locally
Your local AI, connected to the tools you already use
Offline AI is a language model that runs directly on your own device instead of a remote server. You download a model once — then it answers with the internet off, and nothing you type ever leaves your machine.
Notion, GitHub, Google Drive, and more
Any tool on mcp.so works — Notion pages, Google Drive files, GitHub repos, Slack channels, web search. Paste the server config in Settings and it’s available in every chat from that point on.
Tool calls, no AI middleman
Search the web, read a Notion page, query a database — without sending your prompts to OpenAI or Anthropic. Works with any local model you already have installed.
One config, any model
The tools you connect work with every model in Atomic Chat — local or cloud. Switch from Llama to GPT-4o to Gemini without reconfiguring anything.
How MCP connectors work
Connect any tool to your local model — your prompts go directly from your machine to the tool, with no AI provider in between
Connect Notion, GitHub, or any tool in two clicks
Open Settings → MCP Servers → Add. Paste the server config — name, transport type (STDIO, HTTP, or SSE), and URL or command. The tool is active in every chat immediately. No restart, no re-authentication per model.
Type your request. The model picks the right tool.
“Summarize the project brief in my Notion”, “find recent papers on this topic”, “what changed in this repo this week”. You describe what you need — the model selects and calls the relevant tool, then answers in plain language. Works best with models 7B and up.
Your Notion pages don’t go through someone's servers
When you use a local model, requests go from your hardware to the connected service — no AI provider in the middle. OpenAI doesn’t see your Notion pages. Anthropic doesn’t process your database queries. Your data passes only through the services you explicitly connected.
MCP connectors without cloud routing
When you search the web or read a Notion page in ChatGPT, the request goes through OpenAI’s servers. In Atomic Chat with a local model, it goes directly from your machine to the tool — no AI provider in between.
- Tool calls run on your hardware
- Works with local models, no cloud required
- Any MCP server — STDIO, HTTP, SSE
- Same connectors work across all your models
- Free — no subscription, no API key
- All requests route through external servers
- No local MCP pipeline — tool results go to
the cloud - Locked to one provider's models
- $20/month subscription required
MCP connectors you can add today
Any MCP server from mcp.so works. These are some of the most useful ones — and what each one unlocks when you run a local model.
Notion
Search pages, read databases, pull meeting notes. Ask your local model to summarize a project page or find a decision — without pasting text into a chat window.
Google Drive
Read Docs, Sheets, and files from your Drive. Ask the model to extract, compare, or draft based on documents you own — without uploading them to a cloud AI service.
Web search
Run real-time searches from a local model. The query goes from your machine to the search API — no AI provider in the middle. Useful when your model’s training data is out of date.
GitHub
Browse repos, read issues and PRs, search code. Useful for code review or onboarding — ask a local model about your codebase without sharing it with a cloud provider.
Firecrawl
Turn any website into clean, LLM-ready text. Handles JavaScript rendering and paginated content. Ask a local model to extract, summarize, or compare pages you point at.
Supabase
Query your Postgres database, inspect schema, run migrations. Ask in plain language instead of writing SQL. The model connects directly to your project — no data routed through an AI provider.
1000+ more
Sentry, Slack, Stripe, Linear, Cloudflare, Figma, PostgreSQL, Obsidian, Airtable, Google Drive, Jira, Gmail, Redis, Kubernetes, AWS, and anything else with an MCP server.
MCP connection takes three steps
Download & install
Free for macOS, Windows and Linux. No account needed.
Pick a model
Choose from 1000+ models — it downloads to your disk once.
Add your MCP servers
No toggle, no settings. KV cache compression runs from the first message.
FAQ
How MCP works in Atomic Chat, what’s supported, and what happens to your data
MCP (Model Context Protocol) is an open standard for connecting AI models to external tools and data. An MCP server is a small program that exposes specific capabilities, such as web search, file access, or browser control. Atomic Chat acts as the host: you add a server once, and its tools become available to every model you run in the app, local or cloud.
Only when the server itself calls a web service. A filesystem server reads local files and works with no connection at all; a web search server needs internet to reach its API. The model is a separate question. Run a local model with local servers and Atomic Chat operates fully offline, with nothing leaving your machine.
Atomic Chat supports STDIO (a local process the app starts and manages for you), HTTP, and SSE. That covers nearly every server in practice: most community servers run over STDIO on your own machine, while remote hosted servers connect over HTTP or SSE.
No. Anthropic published the MCP standard but sees none of your traffic, and Atomic Chat doesn't proxy or log tool calls; they go straight from your machine to the server you connected. The only exceptions are ones you pick yourself: a cloud model sends prompts to its provider, and a server that wraps a web API sends queries there. With a local model and local servers, nothing leaves your device.
Start with the official modelcontextprotocol/servers repository on GitHub, or browse mcp.so, which lists hundreds of community servers by category. Atomic Chat works with any server that follows the MCP spec, including ones you build yourself.
Built in the open
Follow the project, file issues, and chat with the people building Atomic Chat.
Related features
Private AI
Prompts, files and chats stay on your device — never sent to a server.
TurboQuant
Run models smoothly on the hardware you already own.
Local API endpoint
Point OpenClaw, Hermes or any OpenAI-compatible tool at a local model.
Run AI offline for free
A step-by-step guide to running AI with the internet off — no account.
.svg)
