Simple MCP Server
A minimal MCP server providing tools for time, calculations, sandboxed file operations, and safe URL fetching over stateless HTTP.
README
Simple MCP Server
A small Node.js and TypeScript MCP server that exposes time, calculation, restricted filesystem, and public URL fetching tools over stateless Streamable HTTP.
Requirements
- Node.js 20 or newer
- npm
Setup and start
npm install
npm run build
npm start
The default MCP endpoint is http://127.0.0.1:3000/mcp. A health check is available at http://127.0.0.1:3000/health.
For explicit local configuration:
$env:MCP_WORKSPACE_ROOT = "C:\path\to\allowed\workspace"
$env:MCP_ALLOW_WRITE = "false"
$env:MCP_HTTP_HOST = "127.0.0.1"
$env:MCP_HTTP_PORT = "3000"
npm start
Tools and resources
| Tool | Description |
|---|---|
get_current_time |
Get the current time in an IANA time zone. |
calculate |
Evaluate a side-effect-free mathematical expression. |
read_text_file |
Read a UTF-8 file under the configured workspace. |
list_directory |
List a directory under the configured workspace. |
write_text_file |
Write a UTF-8 file when MCP_ALLOW_WRITE=true. |
fetch_url |
Fetch a public HTTP(S) URL with SSRF, timeout, and size checks. |
Resources are available at server://info and workspace://config.
Configuration
| Variable | Default | Purpose |
|---|---|---|
MCP_WORKSPACE_ROOT |
Current directory | Filesystem sandbox root. |
MCP_ALLOW_WRITE |
false |
Enables write_text_file. |
MCP_MAX_FILE_BYTES |
1048576 |
Maximum file read/write size. |
MCP_FETCH_TIMEOUT_MS |
10000 |
Whole-request timeout, including redirects and body reads. |
MCP_MAX_RESPONSE_BYTES |
2097152 |
Maximum fetched response body size. |
MCP_HTTP_HOST |
127.0.0.1 |
HTTP listen address. |
MCP_HTTP_PORT |
3000 |
HTTP listen port. |
MCP_HTTP_PATH |
/mcp |
MCP endpoint path. |
MCP_API_KEY |
Empty | Bearer token. Required for non-loopback listen addresses. |
SIMPLE_MCP_API_KEY |
Empty | Alias for MCP_API_KEY, matching the Codex client variable. |
MCP_ALLOWED_HOSTS |
Empty | Optional comma-separated Host header allowlist. |
Connect an external agent
For access from another machine, bind the server to all interfaces and configure a strong API key:
$env:MCP_HTTP_HOST = "0.0.0.0"
$env:MCP_HTTP_PORT = "3000"
$env:MCP_API_KEY = "replace-with-a-long-random-secret"
$env:MCP_ALLOWED_HOSTS = "mcp.example.com"
npm start
Configure a URL-based MCP client such as Codex:
[mcp_servers.simple-toolbox]
url = "https://mcp.example.com/mcp"
bearer_token_env_var = "SIMPLE_MCP_API_KEY"
Set SIMPLE_MCP_API_KEY in the Agent process environment to the same value as the server's MCP_API_KEY. For local development, use http://127.0.0.1:3000/mcp; for remote access, put the service behind an HTTPS reverse proxy. Do not expose plain HTTP or enable filesystem writes on an untrusted network.
Development
npm run dev
npm test
npm run test:smoke
npm run test:smoke builds the project, starts it on a random local port, verifies Bearer authentication, completes a Streamable HTTP MCP handshake, lists tools and resources, and calls the calculator tool.
The server is stateless: each MCP POST request creates an isolated server transport. Standalone GET/SSE and DELETE session operations return HTTP 405. This is appropriate for the current request-response tools, but not for server-initiated notifications or resumable long-running streams.
The network guard rejects localhost, private, link-local, multicast, and reserved IP address ranges. DNS rebinding protection at the outbound socket connection layer is outside this simple server's scope; deploy behind an egress proxy when stronger network isolation is required.
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。