Daintree MCP
Daintree runs Claude Code, Codex and other agent CLIs in parallel and exposes control of them through a local MCP server
- 87 ★ on the repository
- Last commit
Broad access
Why this level
- An agent with full tier tools opens worktrees and launches other agents with terminal commands
- The app reads the output of all terminals to detect agent state
- Outside MCP access with an API key opens terminal control
Install
- Works in
Manual install
brew install --cask daintreehq/tap/daintreemacOS through Homebrew. Windows and Linux builds are on daintree.org/download.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- An agent with full tier tools opens worktrees and launches other agents with terminal commands
- The app reads the output of all terminals to detect agent state
- Outside MCP access with an API key opens terminal control
This is third-party code. Review the repository files before installing.
What it does
Daintree is an Electron desktop app: every task gets its own git worktree and its own terminal with an agent. Claude Code, Codex, Antigravity, OpenCode and 13 more CLIs run as they are, with your config and login. The app reads terminal output, the window title and the process tree and shows which run is working, waiting or finished. The built-in MCP server exposes app actions as tools: open a worktree, launch another agent, read a sibling terminal state, wait for it to finish. Outside access is off, and once enabled a client with an API key gets 36 tools for terminals, agents, worktrees, recipes and IDE context. A project sets a core or full tier, and agents launched there get that tier tools. Every call is checked against its tier and written to a local log. A dedicated `ask_human` tool lets an agent ask a question and end its turn: you answer later, and Daintree types the answer into the terminal when the agent is idle. For review there is a Review Hub panel with diffs for each worktree.
Who it is for. Developers who run several agents on one project and want their statuses and diffs in view, and who want one agent to control the others.
Good fit when
- Four agents work on different tasks and you need to see which one waits for your decision
- A coordinator agent has to open worktrees and launch helpers through MCP on its own
- Diffs for each branch must be reviewed before merging and committed in the same place
- Agent questions should reach you from a phone through `ask_human`
Not a fit when
- You need one agent in one terminal, which makes the app excessive
- Agent status detection must be exact: the README calls it a heuristic that can be wrong
- You need an MCP server without a desktop app
Example request
Open a separate worktree for the issue task, launch Codex in it and tell me when it finishesMCP
- Transport
- http
- Authentication
- API key
Limitations
- The MCP server is built into the app and does not run apart from Daintree.
- Outside access is off by default and is enabled in settings, and a client authorizes with an API key.
- Agent status comes from terminal heuristics and may be wrong.
- GitHub did not detect the repository license (NOASSERTION), so read the LICENSE file for the terms.
- Login and limits of the agents stay with the CLIs themselves: Daintree does not proxy them and brings no agents of its own.
How to disable. Turn off outside MCP access in Daintree settings. The app uninstalls like any other: `brew uninstall --cask daintree` on macOS.
FAQ
Which tools does the Daintree MCP server expose?
36 tools for terminals, agents, worktrees, recipes and IDE context. The set depends on the core or full tier of the project.
Which agents are supported?
Claude Code, Codex, Antigravity, OpenCode and 13 more CLIs run without a proxy, with their own config and login.
How do I install Daintree?
On macOS `brew install --cask daintreehq/tap/daintree`, on Windows and Linux an installer, AppImage or deb from daintree.org/download.
Is MCP access on by default?
No, outside access is off. Once enabled, a client with an API key gets the tool set of its tier.
What does the `ask_human` tool do?
An agent asks a question through it and ends its turn. You answer later and Daintree types the answer into the terminal once the agent is idle.
Discussion
No questions or reviews yet.
Related
MCP serversA local control panel for coding agents: it runs Claude Code and Codex, keeps a task queue and merges verified work into main.
DevChain runs a team of AI coding agents on your own machine with a shared board, chat, code review and live terminals.
A tmux-based TUI that runs Claude Code, Codex and other agents in one list: status, quick prompts, worktrees and diff review
PromptOps keeps Claude Code, Codex and Gemini CLI side by side by project and gives a lead agent MCP to launch helpers
A workspace for parallel Claude Code and Codex tasks, with separate Git worktrees, terminals and change review
Claudette runs several Claude Code sessions in parallel, each in its own git worktree, from a window or the console
README in short
The README opens with downloads and a platform table, then covers worktree handling, the agent list, the MCP server section, the Assistant and the Canopy inbox. It ends with the build from source and docs on plugins and architecture.