1mcp-app/agent

1MCP - One MCP Server for All

A unified Model Context Protocol server implementation that aggregates multiple MCP servers into one.

Table of Contents

• Overview
• Features
• Quick Start
• Commands
• Prerequisites
• Usage
• Docker
• Trust Proxy Configuration
• Understanding Tags
• Configuration
• Authentication
• Health Monitoring
• How It Works
• Development
• Contributing
• License

Overview

1MCP (One MCP) is designed to simplify the way you work with AI assistants. Instead of configuring multiple MCP servers for different clients (Claude Desktop, Cherry Studio, Cursor, Roo Code, Claude, etc.), 1MCP provides a single, unified server.

Features

• Unified Interface: Aggregates multiple MCP servers into one.
• Resource Friendly: Reduces system resource usage by eliminating redundant server instances.
• Simplified Configuration: Simplifies configuration management across different AI assistants.
• Standardized Interaction: Provides a standardized way for AI models to interact with external tools and resources.
• Dynamic Configuration: Supports dynamic configuration reloading without server restart.
• Async Loading: Optional asynchronous server loading with real-time capability updates via listChanged notifications.
• Graceful Shutdown: Handles graceful shutdown and resource cleanup.
• Secure: Includes comprehensive authentication and security features.
• Optimized: Supports advanced filtering, pagination, and request optimization.
• Health Monitoring: Built-in health check endpoints for monitoring and observability.

Quick Start

To enable Cursor to use existing MCP servers already configured in Claude Desktop, follow these steps:

1. Run the 1MCP server with the Claude Desktop config file:

npx -y @1mcp/agent --config ~/Library/Application\ Support/Claude/claude_desktop_config.json

2. Add the 1MCP server to your Cursor config file (~/.cursor/mcp.json):

{
  "mcpServers": {
    "1mcp": {
      "type": "http",
      "url": "http://localhost:3050/sse"
    }
  }
}

3. Enjoy it!

Commands

1MCP provides several commands to manage your MCP server configuration:

Core Commands

• npx -y @1mcp/agent [serve] - Start the 1MCP server (default command)
  • --transport - Choose transport type (stdio, http, sse)
  • --config - Use specific config file
  • --port - Change HTTP port

MCP Management

• npx -y @1mcp/agent mcp add <name> - Add a new MCP server to configuration
• npx -y @1mcp/agent mcp remove <name> - Remove an MCP server
• npx -y @1mcp/agent mcp list - List all configured MCP servers
• npx -y @1mcp/agent mcp status [name] - Show server status and details
• npx -y @1mcp/agent mcp enable/disable <name> - Enable or disable servers
• npx -y @1mcp/agent mcp update <name> - Update server configuration

App Integration

• npx -y @1mcp/agent app consolidate - Consolidate configurations from other MCP apps
• npx -y @1mcp/agent app discover - Discover MCP servers from installed applications
• npx -y @1mcp/agent app list - List discovered applications
• npx -y @1mcp/agent app status - Show consolidation status

For detailed command usage, run: 1mcp <command> --help

Full documentation: Commands Reference

Prerequisites

• Node.js (version 21 or higher)
• pnpm

Usage

You can run the server directly using npx:

# Basic usage (starts server with SSE transport)
npx -y @1mcp/agent

# Use existing Claude Desktop config
npx -y @1mcp/agent --config ~/Library/Application\ Support/Claude/claude_desktop_config.json

# Use stdio transport instead of SSE
npx -y @1mcp/agent --transport stdio

# Use external URL for reverse proxy setup (nginx, etc.)
npx -y @1mcp/agent --external-url https://example.com

# Configure trust proxy for reverse proxy setup
npx -y @1mcp/agent --trust-proxy=192.168.1.1

# Show all available options
npx -y @1mcp/agent --help

Available options:

| --enable-async-loading | ONE_MCP_ENABLE_ASYNC_LOADING | Enable asynchronous MCP server loading(boolean) | false | | --health-info-level | ONE_MCP_HEALTH_INFO_LEVEL | Health endpoint information detail level ("full", "basic", "minimal") | "minimal" | | --help, -h | | Show help | |

Docker

You can also run 1MCP using Docker:

# Pull the latest image
docker pull ghcr.io/1mcp-app/agent:latest

# Run with HTTP transport (default)
docker run -p 3050:3050 ghcr.io/1mcp-app/agent

# Run with a custom config file
docker run -p 3050:3050 -v /path/to/config.json:/config.json ghcr.io/1mcp-app/agent --config /config.json

# Run with stdio transport
docker run -i ghcr.io/1mcp-app/agent --transport stdio

Available image tags:

• latest: Latest stable release
• vX.Y.Z: Specific version (e.g. v1.0.0)
• sha-<commit>: Specific commit

Examples:

# Custom port and tags
docker run -p 3051:3051 \
  -e ONE_MCP_PORT=3051 \
  -e ONE_MCP_TAGS=network,filesystem \
  ghcr.io/1mcp-app/agent

# With external URL for reverse proxy
docker run -p 3050:3050 \
  -e ONE_MCP_EXTERNAL_URL=https://mcp.example.com \
  -e ONE_MCP_TRUST_PROXY=true \
  ghcr.io/1mcp-app/agent

Trust Proxy Configuration

When running 1MCP behind a reverse proxy, configure trust proxy settings for proper client IP detection:

# Default (safe for local development)
npx -y @1mcp/agent --trust-proxy=loopback

# Behind reverse proxy
npx -y @1mcp/agent --trust-proxy=192.168.1.1

# Behind CDN/Cloudflare
npx -y @1mcp/agent --trust-proxy=true

See for detailed configuration options, security considerations, and reverse proxy setup examples.

Understanding Tags

Tags help you control which MCP servers are available to different clients. Think of tags as labels that describe what each server can do.

How to Use Tags

1. In your server config: Add tags to each server to describe its capabilities

{
  "mcpServers": {
    "web-server": {
      "command": "uvx",
      "args": ["mcp-server-fetch"],
      "tags": ["network", "web"],
      "disabled": false
    },
    "file-server": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "~/Downloads"],
      "tags": ["filesystem"],
      "disabled": false
    }
  }
}

2. When starting 1MCP in stdio mode: You can filter servers by tags

# Only start servers with the "network" tag
npx -y @1mcp/agent --transport stdio --tags "network"

# Start servers with either "network" or "filesystem" tags
npx -y @1mcp/agent --transport stdio --tags "network,filesystem"

3. When using SSE transport: Clients can request servers with specific tags

{
  "mcpServers": {
    "1mcp": {
      "type": "http",
      "url": "http://localhost:3050/sse?tags=network" // Only connect to network-capable servers
    }
  }
}

Example tags:

• network: For servers that make web requests
• filesystem: For servers that handle file operations
• memory: For servers that provide memory/storage
• shell: For servers that run shell commands
• db: For servers that handle database operations

Configuration

Global Configuration

The server automatically manages configuration in a global location:

• macOS/Linux: ~/.config/1mcp/mcp.json
• Windows: %APPDATA%/1mcp/mcp.json

Configuration File Format

{
  "mcpServers": {
    "mcp-server-fetch": {
      "command": "uvx",
      "args": ["mcp-server-fetch"],
      "disabled": false
    },
    "server-memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"],
      "disabled": false
    }
  }
}

Authentication

1MCP supports OAuth 2.1 for secure authentication. To enable it, use the --enable-auth flag. The --auth flag is deprecated and will be removed in a future version.

When authentication is enabled, 1MCP acts as an OAuth 2.1 provider, allowing client applications to securely connect. This ensures that only authorized clients can access the MCP servers.

Health Monitoring

1MCP provides comprehensive health check endpoints for monitoring and observability:

Health Check Endpoints

• GET /health - Complete health status including system metrics, server status, and configuration
• GET /health/live - Simple liveness probe (always returns 200 if server is running)
• GET /health/ready - Readiness probe (returns 200 if configuration is loaded and ready)

Health Status Levels

• healthy - All systems operational (HTTP 200)
• degraded - Some issues but still functional (HTTP 200)
• unhealthy - Critical issues affecting functionality (HTTP 503)

Monitoring Integration

Use these endpoints with:

• Load balancers (health checks)
• Container orchestration (Kubernetes health probes)
• CI/CD pipelines (deployment validation)

How It Works

1MCP acts as a proxy, managing and aggregating multiple MCP servers. It starts and stops these servers as subprocesses and forwards requests from AI assistants to the appropriate server. This architecture allows for a single point of entry for all MCP traffic, simplifying management and reducing overhead.

System Architecture

graph TB
    subgraph "AI Assistants"
        A1[Claude Desktop]
        A2[Cursor]
        A3[Cherry Studio]
        A4[Roo Code]
    end

    subgraph "1MCP Server"
        MCP[1MCP Agent]
    end

    subgraph "MCP Servers"
        S1[Server 1]
        S2[Server 2]
        S3[Server 3]
    end

    A1 -->|http| MCP
    A2 -->|http| MCP
    A3 -->|http| MCP
    A4 -->|http| MCP

    MCP --> |http| S1
    MCP --> |stdio| S2
    MCP --> |stdio| S3

Request Flow

sequenceDiagram
    participant Client as AI Assistant
    participant 1MCP as 1MCP Server
    participant MCP as MCP Servers

    Client->>1MCP: Send MCP Request
    activate 1MCP

    1MCP->>1MCP: Validate Request
    1MCP->>1MCP: Load Config
    1MCP->>MCP: Forward Request
    activate MCP

    MCP-->>1MCP: Response
    deactivate MCP

    1MCP-->>Client: Forward Response
    deactivate 1MCP

Development

Install dependencies:

pnpm install

Build the server:

pnpm build

For development with auto-rebuild:

pnpm watch

Run the server:

pnpm dev

Debugging

Using the MCP Inspector, which is available as a package script:

pnpm inspector

The Inspector will provide a URL to access debugging tools in your browser.

Debugging & Source Maps

This project uses source-map-support to enhance stack traces. When you run the server, stack traces will reference the original TypeScript source files instead of the compiled JavaScript. This makes debugging much easier, as error locations and line numbers will match your source code.

No extra setup is required—this is enabled by default. If you see a stack trace, it will point to .ts files and the correct line numbers. 🗺️

Contributing

Contributions are welcome! Please read our for details on our code of conduct, and the process for submitting pull requests to us.

License

This project is licensed under the Apache License 2.0. See the file for details.