mcpware

What is mcpware?

mcpware is a gateway that routes MCP requests intelligently to multiple backend servers with comprehensive security validation. It exposes only two routing tools while providing access to an unlimited number of backend tools, bypassing typical MCP client tool limits. Built with Docker-based isolation, it is designed for developers who need to connect to multiple MCP servers through a single entry point.

How to use mcpware?

Install by cloning the repository, building the Docker image, and configuring backends in config.json with environment variables in .env. Add mcpware to Claude Desktop’s claude_desktop_config.json as a Docker-based MCP server, mounting the Docker socket to allow launching backend containers. Once configured, invoke the two main tools: use_tool (route a tool call to a specific backend) and discover_backend_tools (list available backends and their tools).

Key features of mcpware

• Routes MCP requests to multiple backend servers through a single gateway
• Exposes only 2 routing tools, bypassing client-side tool limits
• Experimental security with taint tracking and cross-backend isolation
• Docker-based deployment and automatic backend process management
• Supports environment variable substitution in backend configurations
• Works alongside other MCP servers in the same Claude Desktop setup

Use cases of mcpware

• Connect a single Claude Desktop instance to many MCP backends (e.g., GitHub, databases) without hitting tool count limits
• Organize multiple backend servers behind one gateway while keeping security classification (public/internal/sensitive) for each
• Mix gateway‑managed backends with directly connected MCP servers for specialized workflows
• Gain experimental protection against cross‑backend data leakage and SQL injection

FAQ from mcpware

What is the main benefit of using mcpware instead of multiple direct MCP servers?

mcpware bypasses client-side tool limits by exposing only two routing tools, while giving access to an unlimited number of backend tools through those two tools.

How do I install and configure mcpware?

Clone the repository, build the Docker image, create a config.json with backend definitions and security classification, set environment variables in .env, and add mcpware to your Claude Desktop configuration with the Docker socket mounted.

What are the runtime dependencies?

The production runtime requires Docker. For development, Python 3.10+ and the packages in requirements.txt (pytest, etc.) are needed.

How does the security feature work?

Experimental security prevents cross-backend data leakage (e.g., from a database to GitHub), blocks SQL injection and sensitive data exposure, uses taint tracking to stop all access after suspicious activity, and requires a mandatory security classification (public/internal/sensitive) for each backend.

Does mcpware work on Windows?

Yes, but the Docker socket mount path differs: use //./pipe/docker_engine instead of /var/run/docker.sock. A quick docker version check can verify whether you are using WSL2 (Linux containers) or Windows containers.