The MCP server
For agents that would rather call tools than run shell commands, alto mcp serves the same commands over the Model Context Protocol, on standard input and output.
claude mcp add alto -- alto mcpAny MCP client that starts servers as a command works the same way: the command is alto mcp, with no arguments.
The tools
Section titled “The tools”| Tool | Does |
|---|---|
run |
Runs a command in a terminal you can see and returns its output and exit status as JSON |
wait |
Waits for a state (idle, working, needs-you, done, prompt) or for text to appear |
read_screen |
What the terminal shows now, with the cursor, for full-screen programs |
send_keys |
Types text, then keys by name (Enter, C-c, S-Tab) |
work |
The work panes linked to this conversation |
list_sessions |
Sessions here or on another device, with what each is doing |
use |
Act on a given session from now on, or start one |
Every tool acts on the conversation’s work pane unless it’s given a session (and device), or use chose one. The full arguments are in the MCP tool reference.
Who’s typing
Section titled “Who’s typing”Input from MCP tools is an agent’s, like alto run from an agent: it waits while you type, is refused while the pane is paused, and is marked as typed by the agent.