diffsorted
Splits messy, multi-concern git diffs into focused groups and generates a clear, accurate commit message for each, exposed as an MCP server for use with MCP-compatible clients.
README
diffsorted
Splits messy, multi-concern git diffs into focused groups and generates a clear, accurate commit message for each — exposed as an MCP (Model Context Protocol) server so any MCP-compatible clients (Claude Desktop, Claude Code, etc.) can use it directly against a local repo.
Why
Built as a follow-up to CommitMsg, an AI commit-message generator I shipped earlier. CommitMsg works well on small, single-concern diffs, but struggles when a diff spans multiple unrelated changes (e.g. a bug fix + unrelated formatting pass, or multiple files) — it either writes one vague message or focuses on the wrong part of the diff.
diffsorted solves this by splitting the diff before generating messages, so each commit message is scoped to one actual concern — closer to how atomic commits should look.
How it works
- Parse — reads the raw
git diffoutput and breaks it into per-file segments. - Split — groups related files into concerns using one of two strategies:
splitByFile(v1) — one chunk per changed file.splitByConcern(v2) — groups files by directory/naming (e.g.auth.js+auth.test.js), and separates out formatting-only changes into their own chunk.
- Generate — calls Groq's LLaMA API once per chunk to produce a focused conventional commit message.
Tools exposed
| Tool | Description |
|---|---|
get_diff |
Returns the raw git diff for a repo (staged or unstaged) |
generate_commit_messages |
v1: one commit message per changed file |
generate_commit_messages_v2 |
v2: one commit message per logical concern group |
Setup
```bash git clone https://github.com/bhakti0725/diffsorted.git cd diffsorted npm install ```
Create a .env file:
```
GROQ_API_KEY=your_key_here
```
Get a free key at console.groq.com/keys.
Connecting to Claude Desktop
Add to your claude_desktop_config.json:
```json
{
"mcpServers": {
"diffsorted": {
"command": "npx",
"args": ["tsx", "/absolute/path/to/diffsorted/src/index.ts"]
}
}
}
```
https://www.npmjs.com/package/diffsorted
visit the above link, and download the mcp server as a npm package using npm i diffsorted.
No cloning or manual setup needed — npx fetches and runs it automatically. You'll still need a Groq API key in a .env file in your working directory (see Setup above).
Tech stack
TypeScript, Model Context Protocol SDK, Groq (LLaMA 3.3 70B), Node.js child_process for git integration.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。