TAI42 — the OS for AI tools

Connect any MCP client to a running server over its published transport. The runtime is a standard MCP server, so any MCP client connects to it the same way it connects to any other — over the transport the server publishes. The only choices are which transport the server runs and how the client addresses it.

Match the transport

Start the server on the transport your client speaks (see Deploy):

stdio — the client launches the server as a subprocess and speaks over its standard streams. Configure the client with a command.
HTTP / streamable-http / SSE — the client connects to a URL. Configure the client with the server’s address.

tai serve --transport stdio                          # client launches the server
tai serve --transport http --host 0.0.0.0 --port 8000   # client connects to a URL

Configure the client

Most MCP clients read a JSON config that maps a server name to either a launch command (stdio) or a URL (HTTP).

{
  "mcpServers": {
    "tai": {
      "command": "tai",
      "args": ["serve", "--transport", "stdio"]
    }
  }
}

{
  "mcpServers": {
    "tai": {
      "url": "http://localhost:8000/mcp"
    }
  }
}

A server behind access control requires an API key. Supply it the way your client passes credentials — for an HTTP transport, an Authorization header; for a stdio server, the TAI_API_KEY environment variable on the launched process.

Scope a client to a subset

To expose only a subset of tools to one client, publish them as a sub-MCP app and point the client at its slug. See Expose a private tool set.

Integrations

MCP clients

Match the transport

Configure the client

Scope a client to a subset

See also

​Match the transport

​Configure the client

​Scope a client to a subset

​See also

Match the transport

Configure the client

Scope a client to a subset

See also