Unterm
A terminal emulator that AI agents can drive over MCP, for developers running coding
- Pricing
- Open source
- Platforms
- macOS, Linux, Windows
- Version
- 0.57.4
- Listing updated
- today
Similar MCP servers
About Unterm
Unterm is an open-source MCP server for developer tools. It is available on macOS, Linux, Windows, CLI and self-hosted.
The README says Unterm is a cross-platform terminal built on its own next-core engine, with a local MCP server and CLI so external agents such as Claude Code, Codex or Gemini CLI can spawn shells, run commands, read pane contents and take screenshots. Unlike terminals that embed a chat assistant, it keeps AI generation out of the terminal and leaves that to whichever agent the user connects.
For people running several coding agents at once, it adds an Agent Cockpit: a per-pane status view, an inbox that lists agents waiting for input, and fleets that run one task across several agents in separate git worktrees. A review page shows each member's diff, with squash-merge that stages changes without committing them.
The product is aimed at developers who already use terminal-based coding agents and want to supervise or script them from one place. The site and README describe it as MIT-licensed with no paid tier, and say the MCP and settings servers bind only to 127.0.0.1.
What it does
- Agent state inbox
- Shows which agent in each pane is working, idle, done or waiting for approval, with waiting agents listed first.
- Worktree agent fleets
- Runs one task across several agents, each in its own git worktree and branch beside the main repository.
- Diff review and rollback
- Compares fleet members' diffs side by side and supports squash-merge, discard and rollback from a review page.
- MCP terminal control
- Lets authenticated MCP clients open tabs, run commands, read pane output and capture screenshots over loopback TCP.
- Session recording to markdown
- Records terminal sessions as markdown files with built-in masking of common token patterns.
- Identity profiles
- Binds GitHub, AWS, npm and SSH credentials and git author details to a window, with secrets kept in the OS vault.
Good for
- Watch several Claude Code or Codex sessions and jump to whichever one needs an answer
- Run the same task through two or more coding agents and compare their results before merging
- Let an AI agent open a project tab, run the test suite and read the output back
- Keep separate developer identities for work and personal projects in different terminal windows
Pricing
Open source
MIT-licensed and free; no paid tier stated.
Prices change. Check unterm.app before you buy.
Unterm FAQ
Is Unterm free to use?
The README and website say Unterm is MIT-licensed, with every feature in one binary and no subscription or paid tier.
Which platforms does it run on?
The README lists macOS, Linux and Windows builds, with installers published on GitHub Releases.
How do AI agents connect to it?
Each Unterm window starts a local MCP server on 127.0.0.1 that is protected by an auth token. The setup-ai command registers it with supported agents.
Does Unterm send data to the cloud?
The website states there is no telemetry, analytics, login or cloud round-trip, and that recordings are stored locally.
Is Unterm free?
Yes. Unterm is open source, so you can use and self-host it for free.
What are the best alternatives to Unterm?
The closest alternatives to Unterm are reevesagents, agend-sh, Codeman, agent-manager.dev documentation and Agentic MCP.
Where this comes from
- Description: written from the project's README, the project's website and its MCP Registry entry.