mait.sh1.132.1

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

Install and configure mcp-console

Check that MCP Console is there, make it reachable from your agent, run a first command and keep a shell open for several calls.

Beta
OverviewGet startedUse casesToolsSettingsProblems and fixesAll components
MCP Console

Output varies by version and machine.

Steps

Five steps from install to your first shell.

MCP Console is an MCP tool: you check it is installed, make it reachable, and then ask in plain language. It runs on Linux (amd64, arm64). On macOS and Windows, MCP Console runs in the managed Linux environment that AIDC Launch provides. Local command use needs no remote endpoint; optional browser participation has separate setup.

  1. Capturedmcp-console 0.13.4
    Check the version
    $ mcp-console version
    mcp-console 0.13.4
    Step 1 of 5

    Check that it arrived

    The MAIT image ships mcp-console and updates it with the image. Run mcp-console version to see the version.

    It needs a bash shell and a base64 -d utility on the machine where it runs. See mait-image.

    What you get

    The version command answers, and you know which build you are on when you report a problem.

  2. mcpServers entry
    {
  "mcpServers": {
    "console": {
      "command": "mcp-console",
      "args": ["serve"]
    }
  }
}
    Step 2 of 5

    Make it reachable

    In the MAIT image, Console is available through MCP Registry as console. Confirm the provided connection first. For a separate installation, the service owner can instead supply a direct client launch entry:

    Place the entry where your client keeps server configuration and restart the client. Leave MAIT_LOG_FORMAT unset in the launch environment: its json value corrupts the standard-output stream the client reads.

    What you get

    A client that starts the service on its own, and an agent that can list its tools: terminal_run, terminal_open, terminal_send, terminal_read, terminal_watch, terminal_stats, terminal_list and terminal_close.

  3. Capturedmcp-console 0.13.4
    YouPrint hello-console, then tell me what kind of machine this is.
    Tool callterminal_run { "command": "printf 'hello-console\\n'; uname -sm", "timeout": 10 }
    What happensReturns exit code 0, timedOut false, a duration of about 30 milliseconds and the output hello-console followed by Linux x86_64. The terminal was created for this run and closed after it (ephemeral true).
    AgentIt printed hello-console, and the machine is Linux on x86_64. The command exited with code 0.
    Step 3 of 5

    Run a first command

    Ask your agent for something harmless. It calls terminal_run, which requests a foreground result in a temporary terminal and attempts cleanup afterwards.

    What you get

    A real exit code, the output, and proof that the service can start a shell.

  4. Capturedmcp-console 0.13.4
    YouOpen a shell for this task, move to /tmp and set a variable I can reuse.
    Tool callterminal_open { "name": "guide-demo", "shell": "bash", "size": { "cols": 100, "rows": 30 } }
    What happensAction created: a running bash terminal named guide-demo, 100 columns by 30 rows, with a session id.
    Tool callterminal_send { "session": "<id>", "text": "cd /tmp && export GREETING=kept && echo \"state: $GREETING in $(pwd)\"", "enter": true, "await": { "until": "stable", "ms": 300 } }
    What happensThe send succeeds and the screen change that came back reads: state: kept in /tmp.
    AgentThe shell is open, it is in /tmp and GREETING is set; the next command I send will still see both.
    Step 4 of 5

    Keep a shell open

    When a task needs several steps that share a directory or a variable, ask for a named terminal. The agent opens it once, sends commands into it and reads the result of each.

    What you get

    A terminal with its own state, and a screen read that shows the input was handled, not only delivered.

  5. Capturedmcp-console 0.13.4
    YouI am done with that shell. Close it.
    Tool callterminal_close { "session": "<id>" }
    What happensAction terminated, success true, exit code 0. The exit code of a close is a synthetic value, not the exit status of anything that ran in the shell.
    AgentThe guide-demo shell is closed.
    Step 5 of 5

    Close what you opened

    A named terminal is also subject to the service's connection and idle-retention rules. Ask the agent to clean up when it is done.

    Inspect the per-terminal close result and re-list the intended terminal. If the workload must be confirmed stopped, use its own stop check and ask the service owner to reconcile uncertainty.

    What you get

    A cleanup result for the intended terminal. If its workload must be confirmed stopped, use the workload's own shutdown check and involve the service owner when uncertain.

Result

Commands with real results and shells that remember.

MCP Console is installed, reachable and running commands for your agent, in short runs or in a shell it keeps. Commands run with the permissions of the account that runs the service, so run it where the agent should be allowed to act.

Next

Where to go next.