mcpd-proxy
An MCP server that acts as a proxy between IDEs and the mcpd daemon, exposing all mcpd-managed MCP servers through a unified interface.
README
mcpd-proxy
An MCP (Model Context Protocol) server that acts as a proxy between IDEs and the mcpd daemon, exposing all mcpd-managed MCP servers through a unified interface.
Overview
┌─────────────┐ STDIO/JSON-RPC ┌──────────────┐ HTTP/REST ┌──────────┐
│ IDE/Editor │ ◄─────────────────► │ mcpd-proxy │ ◄───────────────►│ mcpd │
│ (VS Code, │ MCP Protocol │ MCP Server │ Uses mcpd SDK │ daemon │
│ Cursor) │ │ │ │ │
└─────────────┘ └──────────────┘ └──────────┘
mcpd-proxy aggregates tools, resources, and prompts from multiple MCP servers managed by mcpd into a single MCP interface, making it easy for IDEs to access all capabilities without managing individual server connections.
Features
- Unified Interface: Single MCP server exposing all
mcpd-managed capabilities - Tool Aggregation: Tools from all servers with
server__toolnaming convention - Resource Aggregation: Resources from all servers with
server__resourcenaming andmcpd://URIs - Prompt Aggregation: Prompts from all servers with
server__promptnaming convention - Efficient Caching: Leverages SDK caching for health checks and tool schemas
- Zero Configuration: Works out of the box with sensible defaults
- TypeScript: Built with
TypeScriptfor type safety
Prerequisites
Node.js22.10.0 or higher (latest 22.x recommended)mcpddaemon running and accessiblemcpdSDK (automatically installed as a dependency)
Installation
From npm (Recommended)
# Global installation
npm install -g @mozilla-ai/mcpd-proxy
# Or use directly with npx
npx @mozilla-ai/mcpd-proxy
From Source
# Clone the repository
git clone https://github.com/mozilla-ai/mcpd-proxy.git
cd mcpd-proxy
# Install dependencies
npm install
# Build the project
npm run build
Configuration
mcpd-proxy is configured via environment variables:
| Variable | Description | Default |
|---|---|---|
MCPD_ADDR |
mcpd daemon address |
http://localhost:8090 |
MCPD_API_KEY |
Optional API key for mcpd authentication |
(not set) |
Usage
Running Directly
# Using npm package (recommended)
npx @mozilla-ai/mcpd-proxy
# With custom mcpd address
MCPD_ADDR=http://localhost:8090 npx @mozilla-ai/mcpd-proxy
# With API key
MCPD_ADDR=http://localhost:8090 MCPD_API_KEY=your-key npx @mozilla-ai/mcpd-proxy
# From source build
node dist/index.mjs
# From source with custom address
MCPD_ADDR=http://localhost:8090 node dist/index.mjs
VS Code Setup
Add to your VS Code MCP settings file (location varies by platform):
{
"servers": {
"mcpd": {
"type": "stdio",
"command": "npx",
"args": ["@mozilla-ai/mcpd-proxy"],
"env": {
"MCPD_ADDR": "http://localhost:8090"
}
}
}
}
Or if building from source:
{
"servers": {
"mcpd": {
"type": "stdio",
"command": "node",
"args": ["<path-to-mcpd-proxy>/dist/index.mjs"],
"env": {
"MCPD_ADDR": "http://localhost:8090"
}
}
}
}
Replace <path-to-mcpd-proxy> with the absolute path to your installation.
Reload VS Code: Cmd+Shift+P → "Developer: Reload Window"
Verify the connection in the MCP panel to see available tools.
Cursor Setup
Create or edit .cursor/mcp.json in your project directory, or ~/.cursor/mcp.json for global configuration:
{
"mcpServers": {
"mcpd": {
"command": "npx",
"args": ["@mozilla-ai/mcpd-proxy"],
"env": {
"MCPD_ADDR": "http://localhost:8090"
}
}
}
}
Or if building from source:
{
"mcpServers": {
"mcpd": {
"command": "node",
"args": ["<path-to-mcpd-proxy>/dist/index.mjs"],
"env": {
"MCPD_ADDR": "http://localhost:8090"
}
}
}
}
Replace <path-to-mcpd-proxy> with the absolute path to your installation, or use ${workspaceFolder} for relative paths.
Reload Cursor to apply the configuration.
See examples/ folder for configuration examples.
Development
Project Structure
mcpd-proxy/
├── src/
│ ├── index.ts # CLI entry point
│ ├── server.ts # MCP server implementation
│ ├── config.ts # Configuration loader
│ └── apiPaths.ts # API endpoint constants
├── tests/
│ └── unit/ # Unit test files
│ ├── aggregation.test.ts
│ ├── apiPaths.test.ts
│ ├── config.test.ts
│ ├── parsers.test.ts
│ └── server.test.ts
├── .github/
│ └── workflows/ # GitHub Actions workflows
│ ├── tests.yaml
│ ├── lint.yaml
│ └── release.yaml
├── examples/
│ ├── vscode-config.json # VS Code configuration example
│ └── cursor-config.json # Cursor configuration example
├── dist/ # Build output (gitignored)
├── package.json # npm package configuration
├── package-lock.json # npm dependency lock file
├── tsconfig.json # TypeScript compiler configuration
├── tsconfig.test.json # TypeScript test configuration
├── vitest.config.ts # Vitest test configuration
├── vite.config.mts # Vite build configuration
├── eslint.config.mts # ESLint configuration
├── .prettierignore # Prettier ignore patterns
├── .gitignore # Git ignore patterns
└── README.md # This file
Development Workflow
# Install dependencies
npm install
# Build once
npm run build
# Watch mode (auto-rebuild on changes)
npm run dev
# Run tests
npm test
# Run tests in watch mode
npm run test:watch
# Type check without building
npm run typecheck
# Lint code
npm run lint
# Format code
npm run format
Manual Testing
Test the MCP protocol directly using JSON-RPC over stdio:
# Test initialize
echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"1.0.0"}}}' | node dist/index.mjs
# Test list tools
echo '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}' | node dist/index.mjs
Naming Conventions
Tools
Tools are exposed with the format: {server}__{tool_name}
Examples:
time__get_current_time-get_current_timetool fromtimeservergithub__create_issue-create_issuetool fromgithubserverfetch__get_url-get_urltool fromfetchserver
This naming convention prevents tool name collisions between servers and makes it clear which server provides each tool.
Resources
Resources use a custom URI scheme: mcpd://{server}/{resource_uri}
Examples:
mcpd://filesystem/documents/file.txtmcpd://database/users/123
Prompts
Prompts follow the same naming convention as tools: {server}__{prompt_name}
Architecture
Singleton McpdClient
mcpd-proxy creates a single instance of McpdClient at startup and reuses it for all requests. This is essential for:
- Caching: Health check cache (10s
TTL) and tool schema cache (60sTTL) - Performance: Avoids creating new
HTTPconnections for each request - Efficiency: Reduces load on
mcpddaemon
const mcpdClient = new McpdClient({
apiEndpoint: config.mcpdAddr,
apiKey: config.mcpdApiKey,
healthCacheTtl: 10,
});
MCP Protocol Handlers
The proxy implements the following MCP protocol handlers:
initialize- Handshake with IDE, declares capabilitiestools/list- Aggregates tools from allmcpdserverstools/call- Parses tool name and forwards tomcpdresources/list- Aggregates resources from all serversresources/read- Forwards resource read requests tomcpdprompts/list- Aggregates prompts from all serversprompts/get- Forwards prompt requests tomcpdping- Health check endpoint
Troubleshooting
Cannot connect to mcpd daemon
Cause: mcpd daemon is not running or not accessible
Solution:
- Verify
mcpdis running:curl http://localhost:8090/api/v1/servers - Check
MCPD_ADDRenvironment variable is correct - Ensure no firewall blocking the connection
Server not found
Cause: Requested server doesn't exist in mcpd
Solution:
- List available servers:
curl http://localhost:8090/api/v1/servers - Check server is configured in
mcpd - Verify server is healthy:
curl http://localhost:8090/api/v1/health/servers/<server-name>
VS Code not showing tools
Cause: VS Code may not have recognized the MCP server
Solution:
- Check VS Code developer console for errors (Help → Toggle Developer Tools)
- Verify the path to
dist/index.mjsis correct and absolute (if building from source) - Reload VS Code:
Cmd+Shift+P→ "Developer: Reload Window" - Check
mcpddaemon is running and accessible
Tools listed but execution fails
Cause: Server may be unhealthy or tool doesn't exist
Solution:
- Check server health via
mcpdAPI - Verify tool exists on the server
- Check
mcpdlogs for errors
Future Enhancements
- Dynamic tool list updates (
notifications/tools/list_changed) - Server filtering via
MCPD_SERVERSenvironment variable - Improved unhealthy server handling
Related Projects
mcpd- The MCP daemon this proxy connects tomcpd-sdk-javascript-TypeScriptSDK formcpdmcpd-sdk-python- Python SDK formcpd
License
Apache-2.0
Contributing
See the main mcpd repository for contribution guidelines.
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。