1. Conduid
  2. Developer Tools
  3. Claude Code Main
MCP server · Developer Tools

Claude Code Main

Unclaimed last commit 6 months ago devtools
40Fair

Scored 6 months ago · breakdown

About Claude Code Main

Claude Code Main is an MCP server published by somethingforheheda in the Developer Tools category.

The repository has 103 stars and 200 forks, with the last commit 6 months ago. Six months or more without a commit doesn't mean the server is broken, but check the open issues before depending on it in production.

Install

From source
git clone https://github.com/somethingforheheda/claude-code-main

This server has no ConduID identity, so agent calls to it are not receipted. Pin the version you install and review the source before granting it credentials.

Ask AI

✦

Ask AI about Claude Code Main

Powered by Claude · Grounded in docs

I know everything about Claude Code Main. Ask me about installation, configuration, usage, or troubleshooting.

Security checks

  • ✓README presentThe repository ships documentation.

5 checks have not been run yet.

README

Claude Code 源码编译运行指南

环境要求

  • Bun >= 1.3.11
  • Node.js >= 18.0.0

快速开始

1. 安装依赖

bun install

2. 编译

bun run build

编译成功后会在根目录生成 cli.js(约 22MB)。

3. 运行

# 设置 API Key
export ANTHROPIC_API_KEY=your_api_key_here

# 启动交互界面
bun cli.js

# 或直接传入 prompt(非交互模式)
bun cli.js -p "你好"

常用命令

bun cli.js --version       # 查看版本
bun cli.js --help          # 查看帮助
bun cli.js --model <model> # 指定模型

项目说明

本项目为 @anthropic-ai/claude-code v2.1.88 源码,使用 Bun 作为构建工具,基于 TypeScript + React (Ink) 构建的终端 AI 编程助手。

技术栈

技术 用途
Bun 构建工具 + 运行时
TypeScript 主要开发语言
React + Ink 终端 UI 渲染
Zod v4 数据校验
@anthropic-ai/sdk Anthropic API 客户端
@modelcontextprotocol/sdk MCP 协议支持

构建细节

编译命令(完整)

bun build src/entrypoints/cli.tsx --outfile cli.js --target bun \
  --define 'MACRO.VERSION="2.1.88"' \
  --define 'MACRO.BUILD_TIME="2025-01-01T00:00:00Z"' \
  --define 'MACRO.FEEDBACK_CHANNEL="https://github.com/anthropics/claude-code/issues"' \
  --define 'MACRO.ISSUES_EXPLAINER="https://github.com/anthropics/claude-code/issues"' \
  --define 'MACRO.NATIVE_PACKAGE_URL="https://npmjs.com"' \
  --define 'MACRO.PACKAGE_URL="https://npmjs.com"' \
  --define 'MACRO.VERSION_CHANGELOG=""'

MACRO.* 是 Bun 编译时宏,在源码中作为常量使用,必须在构建时注入。

路径别名

项目在 bunfig.toml 中配置了路径别名:

[build]
alias = { "src" = "./src", "react/compiler-runtime" = "react-compiler-runtime" }

源码中 src/xxx 的绝对路径引用通过此别名解析到 ./src/xxx。


私有包存根说明

以下 4 个 Anthropic 内部私有包在 npm 上不公开,已在 node_modules/ 中创建功能存根:

包名 对应功能 影响
@ant/claude-for-chrome-mcp Claude in Chrome 浏览器控制 Chrome 集成不可用
@ant/computer-use-mcp Computer Use 鼠标键盘控制 Computer Use 不可用
@anthropic-ai/mcpb MCP 插件包(.dxt 格式)安装 插件市场不可用
@anthropic-ai/sandbox-runtime 沙箱文件/网络权限隔离 沙箱模式不可用

核心对话、代码编辑、工具调用等主要功能不受影响。


认证方式

方式一:API Key(推荐开发用)

export ANTHROPIC_API_KEY=sk-ant-xxxx
bun cli.js

方式二:OAuth 登录

bun cli.js
# 启动后在界面中选择登录,走 claude.ai 授权流程

方式三:AWS Bedrock

export AWS_ACCESS_KEY_ID=xxx
export AWS_SECRET_ACCESS_KEY=xxx
export AWS_REGION=us-east-1
bun cli.js --model anthropic.claude-3-5-sonnet-20241022-v2:0

修改源码后如何验证

仓库中已包含编译好的 cli.js,可以直接 bun cli.js 运行,不需要重新编译。

如果你修改了源码想验证效果,流程如下:

# 1. 修改源码(src/ 目录下的任意文件)

# 2. 重新编译
bun run build

# 3. 运行验证
bun cli.js

注意:以下模块涉及 Anthropic 内部私有包,修改后无法编译:

  • src/utils/claudeInChrome/ — Chrome 集成
  • src/utils/sandbox/ — 沙箱模式
  • src/skills/bundled/claudeInChrome.ts
  • src/utils/plugins/mcpbHandler.ts

其他绝大部分功能(UI、命令、工具调用、对话逻辑等)均可正常编译验证。


常见问题

Q: 运行后界面卡住? A: 正常现象,Ink (React 终端 UI) 启动需要 1-2 秒,等待即可。

Q: MACRO is not defined 报错? A: 需要用 bun run build 重新编译,直接运行 .ts 源文件不行。

Q: API Error: headers?.get is not a function? A: 升级 zod 到 v4:bun add zod@^4,然后重新 bun run build。

Q: organization has been disabled 报错? A: API Key 所属组织被禁用。如果是环境变量设置的 Key,尝试 unset ANTHROPIC_API_KEY 后用 OAuth 登录。

README copy captured when this server was indexed; it may lag the repository. The original is authoritative.

Questions

About Claude Code Main

How do I install Claude Code Main?

No package has been published for this server, so install it from source: clone the repository and follow its README.

Is Claude Code Main safe to use with an AI agent?

Its trust score is 40 out of 100 (fair). It passes 1 of the 1 static security checks run so far. It has no ConduID identity yet, so agent calls to it are not receipted.

Is Claude Code Main still maintained?

The last commit was 6 months ago. That's long enough that you should check whether the maintainer is responding to issues before depending on it.