Peter Steinberger released MCPorter, a toolkit for calling Model Context Protocol servers from TypeScript or the command line. No hand-written integration code required.
The tool finds MCP servers you've already configured in Cursor, Claude Code, Windsurf, or VS Code. From those definitions, it generates typed clients and CLI wrappers. It handles OAuth, supports HTTP, SSE, and stdio transports, and caches credentials so you're not re-authenticating constantly. This contrasts with older methods that rely on plaintext tokens, a distinction detailed in the MCP vs. Skills comparison.
Version 0.9.0 adds per-server tool filtering and improves stdio shutdown handling and argument coercion. The practical upshot: if you're building automations or AI agents that need external tools, MCPorter removes a lot of glue code. It's already serving as the MCP bridge for OpenClaw, an AI agent framework that relies on it to connect to external tools without bloating its core. Other tools like ClawRun aim to solve related problems for agent developers.
Runs on macOS, Linux, and Windows. Available via npm and Homebrew. Try npx mcporter list to see what MCP servers are sitting on your system right now.
Steinberger built something pragmatic here. No grand framework, just a clean answer to the question: how do I actually call these MCP servers from code?