mait.sh1.69.2

Private beta: features and content may change. No warranty; use at your own risk.

Docs / MCP tool

MCP Console, terminals for your agent

Give an MCP client controlled access to terminals: run a command, keep a shell open, read its screen, and share it with a person.

The terminals carry the authority of the environment they run in.

Your agent sends the calls; you read the results and can join the same terminal.

Basics
First terminal session

What MCP Console does

MCP Console gives an MCP client controlled access to terminal processes in its execution environment. An agent can run a command, keep an interactive shell or REPL open, inspect its screen and output, observe resource usage and close the terminal when it has finished. An optional browser bridge lets a person take part in the same terminal.

The server supplies terminal operations. The host or container supplies the shell, programs, files, permissions and network access those operations use.

Choose a working mode

Your taskUse
A self-contained commandterminal_run without a terminal identifier, then inspect its result
A shell whose directory or variables carry between commandsterminal_open once, then reuse the returned identifier
An interpreter or full-screen applicationterminal_send for input and terminal_read for the resulting screen
Long-running workA background run, checked with screen reads and execution history
A person joining the sessionThe optional browser attachment for one terminal
Your task
A self-contained command
Use
terminal_run without a terminal identifier, then inspect its result
Your task
A shell whose directory or variables carry between commands
Use
terminal_open once, then reuse the returned identifier
Your task
An interpreter or full-screen application
Use
terminal_send for input and terminal_read for the resulting screen
Your task
Long-running work
Use
A background run, checked with screen reads and execution history
Your task
A person joining the session
Use
The optional browser attachment for one terminal

Read the screen or the execution history to confirm a workload finished, and keep required build reports in the workload's own files.

The eight tools

  • terminal_run runs a command in an existing shell, or creates a terminal for it.
  • terminal_open creates, changes or re-attaches a terminal.
  • terminal_send sends text and named keys to a live terminal.
  • terminal_read reads the screen, recent lines, history, a region, screen frames or a stored run capture.
  • terminal_watch subscribes to output or resource events.
  • terminal_stats samples CPU and memory for a terminal's process tree.
  • terminal_list lists the terminals visible to the current session.
  • terminal_close closes, force-closes or detaches terminals.

Continuity and access

Ending a client session can detach persistent terminals while the server stays available, so you can re-attach them later. A supported reload can keep processes running.

Each client session keeps its own terminal records, and a browser link works as a bearer credential. MCP Console is designed for callers trusted with the terminal authority it grants.

What you getYour agent works in real terminals it can keep open and watch, and, where the browser bridge is enabled, you can join the same terminal.

In this section

BasicsWhat it is and your first result

First terminal session
Ask your agent to run one harmless command, keep a shell for several commands, and close it deliberately.

Everyday useCommon tasks, step by step

Everyday tasks
Drive a REPL, supervise a background process, and hand a terminal to a person in a browser.

AdvancedSettings and full reference

Tools and settings reference
Arguments and results for the eight terminal tools, and the settings an installation exposes.

HelpWhen something goes wrong

Problems and fixes
Identify the layer that failed: connection, tool, terminal, command or observation.