Your agent gets real terminals and can work in them like a person.
MCP Console gives an agent terminals in the environment you approve. It can request a command result, keep a shell open between steps, answer an interactive program, and start background work while it performs another task. Check the reported exit code and timeout separately: an unavailable exit code leaves the result unknown.
For a build investigation, the agent can retain one shell and inspect each command before continuing. Long output arrives as a bounded preview; retained captures and history can be read separately while available. Keep required reports in the project because capture limits, eviction and service replacement can remove earlier output. Use the service only with people trusted to control its command environment, and coordinate any persistent handoff or browser participation.
What you gain over a plain shell command.
A tool that only runs one command and returns text leaves the agent guessing whether it worked, losing its place between steps and stuck when a program asks a question. A maintained terminal supports those workflows when the agent observes each result and the installation provides the required lifetime.
A clear success or failure
The agent can distinguish an observed exit result from a timeout or unknown outcome
- Typical alternative
- Reading the output and guessing whether a command worked
- Here
- A foreground result includes captured output, timeout state and an exit code when available
A shell that remembers
The agent works as a person at a keyboard would, one step building on the last
- Typical alternative
- Every command starts in a fresh shell with nothing set
- Here
- The agent keeps one named shell open, and it remembers the folder and settings from step to step
Programs that ask questions
The agent can use an interactive prompt or an installer and check what it answered
- Typical alternative
- Commands that hang when a program waits for an answer
- Here
- The agent can type text and keys into a running program, and wait until it has replied
Output kept in bounds
A bounded preview keeps the conversation focused; save required reports before retention ends
- Typical alternative
- The whole output of a build pasted into the conversation
- Here
- The agent sees the start and the end of a long result and reads the middle only when it needs it
Long jobs in the background
The agent can continue other work while it checks the job deliberately
- Typical alternative
- Waiting on a server or a build, or starting it and losing track of it
- Here
- The agent starts the job, carries on with other work, and returns to see how it is doing and how much it uses
Hand-over between sessions
A long job can outlive the session that started it, as long as the service stays up
- Typical alternative
- A process that dies with the session that started it
- Here
- A terminal can be left running and picked up again by the next session
Available through the configured MAIT connection.
MCP Console comes with the MAIT image and updates with it (mait-image). In the MAIT image it is available through MCP Registry as console; confirm the connection supplied by your installation.
Runs on Linux (amd64, arm64). On macOS and Windows, MCP Console runs in the managed Linux environment that AIDC Launch provides. Provide Bash, a compatible base64 -d, and the programs each task needs in the service environment. Commands use the service account's files, credentials and network access. Local command use needs no outbound service, but the workload may use a network. Optional browser sharing needs an approved access route. See Settings before configuring a service or relying on continuity across reloads.
Where it sits in the MAIT toolkit.
Where to read on.
- Get started, Use cases and Tools
- Settings, for the person who runs the service
- Problems and fixes