claude-code-tools
MCP server providing code review, auto-fix, test running, linting, formatting, and refactor suggestions for Claude Code.
README
claude-code-tools
TypeScript toolkit for Claude Code — code review, auto-fix, tests, lint, and format exposed as an MCP server and a small CLI.
What is this?
claude-code-tools is a TypeScript package that implements common developer workflows so Claude (via Model Context Protocol) or your own scripts can:
- Review files or diffs for risky patterns (e.g.
eval,debugger, possible secrets, TODOs) plus optional custom rules - Auto-fix with ESLint
--fix, Prettier, and safe line-based cleanups (standaloneconsole.log/console.debug/console.info, up tomaxFixesPerFile), with an approval gate by default - Run tests via your configured command (default
npm test) - Generate test skeletons for Vitest or Jest
- Lint and format with the ESLint and Prettier APIs
- Suggest refactors using lightweight heuristics (line length, nesting, etc.)
All processing runs locally on your machine.
Requirements
- Node.js 20+ (22+ recommended for tooling parity)
- Optional: Claude Code or any MCP-capable client
Installation
cd claude-code-tools
npm install
npm run build
The dist/ output is gitignored. Run npm run build before using node dist/..., npm run review, npm run fix, or registering dist/index.js in your MCP client.
npm scripts
| Script | Description |
|---|---|
npm run build |
Compile src/ → dist/ |
npm run dev |
Run MCP server on stdio from TypeScript (stdio; blocks) |
npm run lint |
ESLint on src/ and tests/ |
npm test |
Vitest (unit + integration) |
npm run test:integration |
Integration smoke tests |
npm run review |
node dist/cli.js review (after build) |
npm run test-run |
node dist/cli.js test-run |
npm run fix |
node dist/cli.js fix |
npm run test:watch |
Vitest in watch mode |
npm run prepack |
Runs before npm pack / publish (npm run build) |
Quick start
1. MCP server (stdio)
Build, then register the server (absolute path required):
{
"mcpServers": {
"dev-tools": {
"command": "node",
"args": ["/absolute/path/to/claude-code-tools/dist/index.js"]
}
}
}
Restart the client. Registered tools:
| Tool | Purpose |
|---|---|
code_review |
Review a file or diff |
auto_fix |
ESLint/Prettier + safe console-line removals |
test_runner |
Run the test command |
test_writer |
Emit a test skeleton (write optional: save next to source) |
lint |
ESLint |
format |
Prettier |
refactor_suggest |
Heuristic suggestions |
Development (TypeScript, stdio):
npm run dev
2. CLI
After npm run build:
# Review a file (JSON output)
node dist/cli.js review --file src/index.ts
# Only errors and warnings (drop info-level findings)
node dist/cli.js review --file src/index.ts --severity-warn-only
# Pipe a diff
git diff | node dist/cli.js review --diff -
# Run tests (uses `.claude-tools.config.json` when present)
node dist/cli.js test-run
# Auto-fix: needs --auto-approve or CLAUDE_TOOLS_AUTO_APPROVE unless disabled in config
node dist/cli.js fix --file src/buggy.ts --auto-approve
Exit codes (fix): 0 success, 2 file not found, 3 approval required.
After npm link or publishing: claude-code-tools review --file ... (see package.json bin).
3. Programmatic API
import {
reviewCode,
autoFix,
runTests,
writeTests,
runLint,
runFormat,
refactorSuggest,
} from 'claude-code-tools';
const review = await reviewCode({ path: './src/index.ts' });
console.log(review.summary, review.issues);
const fixed = await autoFix({ path: './src/buggy.ts', autoApprove: false });
const tests = await runTests({ command: 'npm test' });
const generated = await writeTests({
path: './src/auth.ts',
framework: 'vitest',
write: false,
});
console.log(generated.generatedCode);
Use write: true to create a *.test.ts / *.spec.ts file next to the source (same as the MCP test_writer tool with write: true).
4. Custom rules
import { addCustomRule } from 'claude-code-tools/rules';
addCustomRule({
name: 'no-console',
description: 'Disallow console.log in production code',
severity: 'warning',
check: ({ source, filePath }) => {
if (!/\.[jt]sx?$/.test(filePath)) {
return [];
}
return source.includes('console.log')
? [
{
rule: 'no-console',
message: 'Avoid console.log in production paths',
severity: 'warning',
suggestion: 'Use a logger',
},
]
: [];
},
});
Configuration
Create .claude-tools.config.json in the project root:
{
"review": {
"ignorePatterns": ["**/*.test.ts", "dist/**"],
"severityThreshold": "warning"
},
"autoFix": {
"requireApproval": true,
"maxFixesPerFile": 10
},
"testRunner": {
"defaultCommand": "npm run test:ci",
"timeoutMs": 60000
},
"lint": {
"eslintConfig": "eslint.config.js"
}
}
severityThreshold:error|warning|info— caps which severities are reported after scanning.autoFix.requireApproval: whenfalse,auto_fixapplies ESLint/Prettier/console cleanups without the approval gate (default istrue).maxFixesPerFile: max number of standaloneconsole.*lines removed per auto-fix run (default10).CLAUDE_TOOLS_AUTO_APPROVE: set to1,true, oryesto allow destructive fixes without per-callautoApprove(see.env.example).
Docker
A Dockerfile is included. From the repo root:
docker build -t claude-code-tools .
docker run -i --rm -v "$(pwd)":/workspace -w /workspace claude-code-tools
Testing
npm run lint
npm test
npm run test:integration
npm test runs the full Vitest suite, including tests/integration/. Use npm run test:integration if you only want the smoke tests.
Project layout
claude-code-tools/
├── src/
│ ├── tools/ # code-review, auto-fix, test-runner, …
│ ├── mcp-server.ts # MCP tool registration
│ ├── cli.ts # CLI entry (also package bin)
│ ├── rules.ts # Custom rules registry (export: claude-code-tools/rules)
│ ├── config.ts
│ └── index.ts # Library API; MCP stdio when executed as main
├── tests/
├── Dockerfile
├── .dockerignore
├── .env.example
├── .gitignore
├── eslint.config.js
├── LICENSE
├── README.md
├── package.json
├── package-lock.json
├── tsconfig.json
└── vitest.config.ts
License
This project is licensed under the MIT License.
Acknowledgements
Built with TypeScript, @modelcontextprotocol/sdk, ESLint, and Prettier.
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。