MiniCode CLI
MiniCode is a compact TypeScript terminal coding agent like Claude Code: tool loop, TUI, skills, MCP and sub-agents, built for study and tinkering
- 1,152 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent writes files and runs commands in your project, although edits can be approved before writing
- The API token is stored in ~/.mini-code/settings.json
- Connected MCP servers and web requests widen the agent's access
Install
Manual install
git clone https://github.com/LiuMengxuan04/MiniCode.git
cd MiniCode
npm install
npm run install-localThe installer asks for the model, ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN. Start afterwards with minicode.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent writes files and runs commands in your project, although edits can be approved before writing
- The API token is stored in ~/.mini-code/settings.json
- Connected MCP servers and web requests widen the agent's access
This is third-party code. Review the repository files before installing.
What it does
MiniCode is a small implementation of a terminal coding agent. A request runs through a model, tool, model loop: the agent reads the workspace, calls tools for files, search, edits, commands and web requests, and replies in the same session. Before a file is written the edit is shown with path and command permission checks. The full-screen interface has input history, transcript scrolling, a slash-command menu and approvals. /plan shows a todo list, /goal carries a goal across turns, /loop repeats a prompt with a delay, and up to three sub-agents only read code while the root agent makes every edit. Sessions are saved per project and can be resumed, renamed, forked and compacted. Skills are found through SKILL.md, and MCP connects over stdio and remote HTTP. The installer asks for the model, ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN and stores settings in ~/.mini-code. The author calls the project a learning tool: it is easy to read and modify. Rust, Python, Go and Java versions are maintained by separate authors.
Who it is for. For developers who want to understand how a Claude Code-style agent is built and make their own tool from it.
Good fit when
- You want a small source of the model, tool, model loop to study and adapt
- You want to try a terminal agent with sub-agents, /goal and /loop without a large platform
- The agent should use an Anthropic-compatible address that you set in ANTHROPIC_BASE_URL
- You need to see an edit and approve it before it is written to a file
- You want to run the interface without calling a model: there is a mock mode
Not a fit when
- You need a finished product agent with support and frequent releases: the author calls the project educational
- Installation must go through a package manager: the repository offers only a clone and npm run install-local
- You need providers beyond an Anthropic-compatible API: the installer asks for exactly such an address and token
- You plan several different agents writing code: MiniCode sub-agents only read, and only the root agent writes
Example request
Start minicode, read the src folder and propose an edit to the config parsing function, showing the diff before writingLimitations
- Installation only from a clone: git clone, npm install and npm run install-local. An npm package is not announced.
- The installer asks for the model, ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN, so you need access to an Anthropic-compatible API.
- Settings and the MCP server list live in ~/.mini-code/settings.json and mcp.json, and MINI_CODE_HOME changes the directory.
- The Rust, Python, Go and Java versions live in separate repositories with their own authors.
- The project describes itself as educational and experimental.
How to disable. Remove the minicode launcher from the folder set by MINI_CODE_BIN_DIR and the ~/.mini-code settings directory. The cloned repository folder can be deleted by hand.
FAQ
How do I install MiniCode?
Clone the repository, run npm install and npm run install-local. The installer asks for the model name, ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN and creates the minicode launcher.
How do I run MiniCode without model access?
Mock mode starts with MINI_CODE_MODEL_MODE=mock npm run dev. It is for checking the interface and the tool loop; it returns no model answers.
What can MiniCode sub-agents do?
Up to three sub-agents run at once and only read code. The root agent makes every change and can wait for sub-agents or close them.
How is MiniCode different from Claude Code?
MiniCode follows the workflow and architectural ideas of Claude Code in far less code, which suits study and experiments. It has no product support.
How do I connect MCP and skills in MiniCode?
Skills are picked up from SKILL.md and listed with /skills. MCP servers connect over stdio or remote HTTP, /mcp shows their status and the list lives in ~/.mini-code/mcp.json.
Discussion
No questions or reviews yet.
Related
CLI toolsA Python terminal coding agent in the manner of Claude Code: reads the repository, edits files, runs commands, with memory, sub-agents and MCP
A terminal coding agent tuned for the deepseek-v4 model: deep thinking, reasoning effort control, Agent Skills and MCP
An open-source TypeScript terminal coding agent tuned for DeepSeek models and cheap caching
An MCP server and CLI for code research: search and understand your project and all of GitHub with fewer tokens
A plugin and skill that makes the agent write minimal code: reuse, stdlib and native platform features come first
Local code graph for agents: an MCP server returns entry points, symbol links and change impact in one call
Comparison with similar tools
Entries that share the main search query, and how they differ.
README in short
The description opens by calling it a lightweight terminal assistant with a Claude Code-like workflow. It continues with a sponsor block, a list of core authors and language versions, reasons to choose the project and a capability list: tool loop, sub-agents, plan, goal, loop, interface, sessions, context accounting, skills and MCP. It ends with installation, launch, mock mode, main commands and links to the docs.