How it works

From installation to your first task.

MCPtoAI has three parts: the web workspace, a secure relay and the Device Agent on your computer. Here is how they work together.

Set up in five steps.

On macOS and Windows, setup happens in the Desktop app and browser. On Linux, install the CLI from PyPI or run the official Docker image, then pair the device from the terminal.

  1. 1

    Choose your platform

    Install MCPtoAI Desktop on macOS or Windows, install the Linux CLI from PyPI, or run the official Docker image on Linux.

  2. 2

    Install the Device Agent

    On macOS and Windows, Desktop installs the bundled Device Agent. On Linux, the CLI or Docker image provides the device-side agent. This is the component that executes local tools and connects the device to MCPtoAI.

  3. 3

    Connect your account

    Sign in through your browser to pair the device with your MCPtoAI account. A unique cryptographic identity is created for the device.

  4. 4

    Add an AI provider

    Enter an API key for a cloud provider, or use a local model. Provider credentials stay on the device in the supported operating-system credential store; they are not uploaded to MCPtoAI servers.

  5. 5

    Work from anywhere

    Choose where conversation history is stored — MCPtoAI Cloud for cross-device access, or This device only for local storage — then open MCPtoAI Web or the local Desktop chat, choose a device and model, and start working.

What happens when you send a message.

  1. 1

    You send a message

    from MCPtoAI Web or MCPtoAI Desktop.

  2. 2

    The relay delivers it

    to your paired device over TLS. MCPtoAI does not currently provide end-to-end encryption, so chat traffic passes through the hosted relay in transit.

  3. 3

    Your device calls the AI

    with your own provider key, stored on the device.

  4. 4

    Tools run under your rules

    Off, Ask or Allow decides what may happen.

  5. 5

    You get the result

    in the same conversation, on any device.

What runs where.

MCPtoAI keeps provider credentials, device private keys, MCP secrets, permission settings and local files on the device. The hosted service provides account, web and relay functions, and stores conversation history only when you choose MCPtoAI Cloud history.

On your computer

  • MCPtoAI Desktop and Device Agent on macOS / Windows, or the Linux agent / Docker image
  • Your AI provider API keys
  • Local models (Ollama, LM Studio)
  • Your files, applications and permission settings
  • Conversation history when you choose This device only

In the MCPtoAI cloud

  • MCPtoAI Web
  • The secure relay
  • Your account and paired devices
  • Conversation history only when you choose MCPtoAI Cloud history

With services you choose

  • The AI providers you add
  • The MCP services you connect
  • Each under your own account and terms

You decide what AI may do.

Every capability of your computer and every MCP tool has one of three levels.

Off

The AI cannot use this capability or tool.

Ask

The AI asks first. Nothing happens until you approve it in the conversation.

Allow

The AI may use it without asking — useful for low-risk actions such as listing files.

Some actions always ask

Writing files, opening applications or links, capturing the screen, camera or microphone, and running commands always require your approval, whatever the setting. Running commands is turned off by default.

Ready to try it?

Install MCPtoAI on your platform and connect your first device in a few minutes.

Explore what MCPtoAI can do

A practical AI workspace, not just a chat window. Connect your computer, supported AI providers and external tools; keep control of the actions they take.

Work with your computer

Use the Device Agent to inspect permitted folders, read files and perform supported tasks. Terminal access is optional and controlled by device permissions.

Bring services together

Connect GitHub, Google Drive and compatible MCP servers. Discover available tools, then choose which the AI can use.

Reuse workflows with Skills

Inspect supported GitHub-hosted Skill packages pinned to a commit, review their files and explicitly approve installation.

Choose the model for the task

Use supported cloud providers or local models, switch within a conversation and retain the available context. Model capabilities and limits vary.

Stay in control

Configure Off, Ask or Allow for individual tools. Provider credentials remain on the device; remote services receive data needed for approved requests.

Understand where data goes

Web messages pass through the hosted TLS relay; this is not end-to-end encryption. Choose cloud or device-only history, and review third-party tool and Skill risks.

Explore Skills