Cunzhi (寸止) MCP
寸止 (Cunzhi)
MCP server in Rust and Tauri: the agent shows a dialog with options before ending a reply, keeps project rules and searches code
- 1,391 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Code search uploads fragments of your project to an external ACE service at the address and token you set
- Stores project rules and preferences in local memory and feeds them into the agent's work
- Installs as executables from GitHub archives and runs as a persistent process of the MCP client
Install
- Works in
Manual install
brew tap imhuso/cunzhi && brew install cunzhiFor macOS. On Windows and Linux download the archive from the releases page and add `寸止` and `等一下` to PATH.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Code search uploads fragments of your project to an external ACE service at the address and token you set
- Stores project rules and preferences in local memory and feeds them into the agent's work
- Installs as executables from GitHub archives and runs as a persistent process of the MCP client
This is third-party code. Review the repository files before installing.
What it does
Cunzhi is an MCP server with a graphical window for working alongside an AI agent. When the agent is about to end the conversation with a line like asking whether you need more help, a dialog appears instead. It offers preset reply options, Markdown input and several ways to enter text. You keep talking in the same dialog until the task is solved. The second function is per-project memory. It stores development rules and your preferences, which the agent pulls into later sessions. The third is a code search tool built on ACE (Augment Context Engine): the agent queries the project in natural language and gets relevant fragments. The install gives two executables. The `寸止` command runs as the MCP server, and `等一下` opens the settings window with feature switches, themes, memory management and a generator for a reference prompt that you add to the agent. The project ships as a Tauri app with prebuilt archives for Linux, macOS and Windows.
Who it is for. For developers working in Claude Desktop, Cursor or another MCP client who want the agent to ask for confirmation in a dialog before it wraps up a task.
Good fit when
- The agent often ends a reply asking whether you need anything else before the task is done, and you want to answer in a dialog
- You want project rules and preferences pulled into each new session without re-explaining
- The agent needs natural-language search over a large codebase through your own ACE service
- You want a settings window to switch features on and generate a prompt for the agent
Not a fit when
- The agent runs unattended, for example in CI: the dialog would wait for an answer nobody gives
- You work on a server without a graphical shell: the app shows a Tauri window
- Code must not go to an outside service: the code search uploads project fragments to the ACE address you set
Example request
Before you finish, show me the dialog with options for next steps and save the code formatting rules to project memoryMCP
- Transport
- stdio
- Authentication
- not required
Limitations
- The repository's last commit is dated May 2026, and the last release, v0.4.0, came out in November 2025.
- The documentation is in Chinese, and so are the interface and hints. The README has no translations.
- Code search needs your own ACE service: settings take an API address and a token. The project provides no such service, and without the address and token the tool stays off.
- For the agent to actually show the dialog, you must add the prompt from settings to its instructions. Installing the server alone is not enough.
How to disable. Remove the `寸止` entry from the client's MCP configuration, for example `claude_desktop_config.json`. On macOS use `brew uninstall cunzhi`; otherwise delete the `寸止` and `等一下` executables from PATH.
FAQ
How do I connect Cunzhi to Claude Desktop?
Install the program and add a server `寸止` with the command `寸止` to `claude_desktop_config.json`. Then run `等一下`, copy the prompt from the "参考提示词" tab and add it to the agent's instructions.
What does code search send to the ACE service?
Only changed blocks. Files are split into 800-line blocks, SHA-256 hashes reveal the changes, and files matched by `.gitignore` are skipped.
Do I need a separate service for code search?
Yes, you need your own ACE service: the tool settings take its API address and a token.
Which platforms are supported?
Linux, macOS on Intel and Apple silicon, and Windows. macOS installs through Homebrew and the others use archives from the releases page.
Is the project still maintained?
The last commit is dated May 2026 and the last release came out in November 2025. Check the open issues in the repository before adopting it.
Discussion
No questions or reviews yet.
Related
MCP serversA local MCP server that lets the agent ask you questions in the console and notify you when a task is done
An MCP server and CLI that find in Claude Code and Codex sessions why a file became what it is: who asked the agent what, when and what changed
A self-hosted task manager for human and agent collaboration through MCP
A local assistant with an MCP bridge: Claude Code and Cursor search its memory, create tasks, send notifications and ask it questions
A desktop pet and menu-bar monitor that shows which coding agent is working, done or waiting for your input
Jira MCP in Go: an agent searches issues by JQL, runs sprints, comments and worklogs and sees an issue's branches and pull requests on Cloud and Server
README in short
The Chinese README explains the project's purpose, shows screenshots of the dialog and settings, gives two install routes, Homebrew for macOS and a manual archive download, and then a four-step setup. A separate section covers the code search tool and thanks the author of acemcp, whose logic was rewritten in Rust.