LimiChan-2026/mewcode

MewCode

A configurable terminal AI coding assistant built with Java.

简体中文 · Contributing · Security · Support

MewCode is an interactive command-line assistant for reading, searching, editing, and reasoning about code in the current working directory. It supports OpenAI-compatible Chat Completions APIs, Anthropic Messages APIs, MCP tools, project trust controls, hooks, persistent sessions, and multi-agent tasks.

Features

• Interactive terminal UI and one-shot prompt mode
• OpenAI, OpenAI-compatible, and Anthropic provider protocols
• Built-in file, search, edit, write, and command-execution tools
• MCP server integration, hooks, skills, session recovery, and context compaction
• Project trust confirmation, permission modes, secret redaction, and Git worktree support
• Background tasks and team collaboration

Requirements

• JDK 21 or later
• Network access to your chosen model provider
• A model provider baseurl, apikey, and model
• Git (recommended; required for Git worktree features)

Check the installed Java version:

java -version

The major version should be 21 or newer.

Quick start

Clone the repository and enter it:

git clone https://github.com/LimiChan-2026/mewcode.git
cd mewcode

Create .mewcode/config.yaml in the project root. This file is intentionally ignored by Git so that your credentials stay local.

OpenAI or compatible API

providers:
  - name: MyOpenAI
    protocol: openai
    # Use the API root. MewCode appends /chat/completions automatically.
    base_url: https://api.openai.com/v1
    api_key: REPLACE_WITH_YOUR_API_KEY
    model: REPLACE_WITH_YOUR_MODEL
    thinking: false

For a third-party compatible service, replace only baseurl, apikey, and model. openai-compat is also accepted as the protocol name.