mcp-devenv
An MCP server that gives your AI assistant full awareness of your local dev environment — running processes, Docker containers, git state, open ports, log files, and more.
README
iwomm-mcp
An MCP server that gives your AI assistant full awareness of your local dev environment — running processes, Docker containers, git state, open ports, log files, and more.
Stop copy-pasting context. Let the AI ask for what it needs.

Tools
| Tool | What it answers |
|---|---|
get_running_processes |
What Node/Python/Go processes are running? What's on port 3000? |
get_docker_state |
Which containers are up? Are they healthy? What are their ports? |
get_git_status |
What branch am I on? What's dirty? Am I ahead/behind? |
get_env_summary |
What keys does my .env have? (Values masked by default) |
get_config_summary |
What's in my application.yml? (.yml, .properties, .env) |
get_dependencies |
What libraries am I using? (pom.xml, build.gradle, go.mod, package.json) |
get_open_ports |
What's listening on this machine right now? |
get_recent_logs |
What did my API service log in the last 50 lines? |
get_project_structure |
What does this codebase look like? (gitignore-aware) |
check_service_health |
Is http://localhost:3000/health responding? How fast? |
Language support
Most tools operate at the OS or network level and work regardless of stack. The table below calls out the gaps.
| Tool | Node / TS | Java (Spring) | Go | Python | Ruby / .NET / Rust |
|---|---|---|---|---|---|
get_project_structure |
✅ | ✅ | ✅ | ✅ | ✅ |
get_git_status |
✅ | ✅ | ✅ | ✅ | ✅ |
get_docker_state |
✅ | ✅ | ✅ | ✅ | ✅ |
get_open_ports |
✅ | ✅ | ✅ | ✅ | ✅ |
check_service_health |
✅ | ✅ | ✅ | ✅ | ✅ |
get_recent_logs |
✅ | ✅ | ✅ | ✅ | ✅ |
get_running_processes |
✅ | ⚠️ all JVM processes share the name java — use filter_name=java + filter_args=myapp.jar |
✅ | ✅ | ✅ |
get_env_summary |
✅ | ⚠️ .env files only — use get_config_summary for application.* |
✅ | ✅ | ✅ |
get_config_summary |
✅ | ✅ .yml · .properties |
❌ | ❌ | ❌ |
get_dependencies |
✅ package.json |
✅ pom.xml · build.gradle[.kts] |
✅ go.mod |
❌ | ❌ |
What still needs copy-paste
Some things Claude can't read yet — paste these into chat when needed:
- Python deps:
requirements.txt,pyproject.toml,setup.py - Rust deps:
Cargo.toml - Ruby deps:
Gemfile - .NET deps:
*.csproj - Python/Rust config:
config.toml,settings.pyand similar formats - Java stack traces in logs:
get_recent_logsreturns raw lines — a trace split across the tail boundary may arrive incomplete - Spring OAuth2 properties: keys containing
auth(e.g.authorization-grant-type) are masked by the secret-detection heuristic; paste the value if needed
Installation
Via npm (recommended)
npm install -g iwomm-mcp
Or use it without installing — npx will fetch and run it on demand (see Claude config below).
From source
git clone https://github.com/dicoy/iwomm-mcp.git
cd iwomm-mcp
npm install
npm run build
Add to Claude Code
claude mcp add devenv -- npx -y iwomm-mcp
Add to Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json on macOS:
{
"mcpServers": {
"devenv": {
"command": "npx",
"args": ["-y", "iwomm-mcp"]
}
}
}
Restart Claude. You should see devenv in /mcp.
Optional: .mcp-context.yml
Drop this file at the root of any project to tell iwomm-mcp about your services:
# .mcp-context.yml
services:
api:
port: 3000
logs: ./logs/api.log
health_endpoint: /health
worker:
logs: ./logs/worker.log
postgres:
port: 5432
env_files:
- .env
- .env.local
# Keys matching these patterns will have their values revealed
# (even if they look like secrets). Supports exact match and * prefix glob.
reveal_env_patterns:
- NODE_ENV
- APP_*
- FEATURE_*
Without this file, get_recent_logs works only with explicit file paths. All other tools work without any config.
Architecture
┌─────────────────────────────────────────────────────────┐
│ Claude (AI) │
└───────────────────────┬─────────────────────────────────┘
│ MCP protocol (stdio)
┌───────────────────────▼─────────────────────────────────┐
│ iwomm-mcp │
│ │
│ ┌──────────────────┐ ┌──────────────────────────┐ │
│ │ Tool Registry │ │ Providers │ │
│ │ │ │ │ │
│ │ get_git_status ─┼────▶│ IGitProvider │ │
│ │ get_docker_* ─┼────▶│ IDockerProvider │ │
│ │ get_processes ─┼────▶│ IProcessProvider │ │
│ │ ... │ │ IEnvProvider │ │
│ └──────────────────┘ │ ILogProvider │ │
│ │ IPortProvider │ │
│ ┌──────────────────┐ └──────────────┬───────────┘ │
│ │ .mcp-context │ │ │
│ │ .yml │ concrete impls │
│ └──────────────────┘ │ │
└───────────────────────────────────────────┼─────────────┘
│
┌─────────────┬───────────────┼──────────────┐
▼ ▼ ▼ ▼
ps/lsof simple-git dockerode fs/readline
Design principles:
- Provider pattern — every tool depends on an interface, not a concrete implementation. Tests mock the interface; the OS is never touched in tests.
- One Zod schema per tool — the same schema drives both MCP input validation and TypeScript types. No duplication.
- Typed error hierarchy —
DockerNotAvailableError,GitNotARepositoryError, etc. Callers can catch exactly what they expect. - Composition root —
createServer()inserver.tsis the only place that wires interfaces to implementations. Everything else is pure.
Development
npm run dev # build in watch mode
npm run typecheck # tsc --noEmit
npm run lint # biome check
npm run lint:fix # biome check --write
npm run test # vitest run
npm run test:watch # vitest (interactive)
npm run ci # typecheck + lint + test + build
Adding a new tool
- Create
src/tools/your-tool-name/schema.ts— Zod input schema - Create
src/tools/your-tool-name/handler.ts— pure function, injected providers - Create
src/tools/your-tool-name/handler.test.ts— mock the providers, not the OS - Register in
src/registry/tool-registry.ts— oneserver.tool(...)call
If the tool needs a new system capability, add an interface to src/providers/ with a corresponding implementation. The handler never imports a concrete provider class.
Project structure
src/
├── config/
│ ├── loader.ts # .mcp-context.yml parser
│ └── schema.ts # Zod schema + inferred types
├── errors/
│ └── index.ts # typed error hierarchy
├── providers/
│ ├── docker.ts # IDockerProvider + DockerodeProvider
│ ├── env.ts # IEnvProvider + FsEnvProvider
│ ├── git.ts # IGitProvider + SimpleGitProvider
│ ├── log.ts # ILogProvider + FsLogProvider
│ ├── port.ts # IPortProvider + LsofPortProvider
│ └── process.ts # IProcessProvider + NodeProcessProvider
├── registry/
│ └── tool-registry.ts # wires tools to the MCP server
├── tools/
│ └── <tool-name>/
│ ├── schema.ts
│ ├── handler.ts
│ └── handler.test.ts
├── index.ts # stdio transport + startup
└── server.ts # createServer() factory
Tech stack
| Runtime | Node.js 20+ |
| MCP SDK | @modelcontextprotocol/sdk |
| Validation | zod |
| Git | simple-git |
| Docker | dockerode |
| Shell | execa |
| Build | tsup |
| Tests | vitest |
| Lint + format | biome |
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。