perceivingai/portus-mcp

Portus MCP is a server that lets AI assistants like ChatGPT and Codex reach out and actually work on code on your local machine or on remote virtual machines, instead of just generating text in a chat window. MCP stands for Model Context Protocol, a standard way for AI tools to connect to external capabilities. Portus MCP acts as that bridge. You run it on any machine you want an AI to be able to access, register the project folders you want it to work in, and then point your AI client at the server's address. From that point the AI can read and write files inside those projects, run commands, and perform real development work directly on your machine. You can run Portus MCP on multiple machines at the same time, for example your local PC, a Linux virtual machine, and a remote workstation, and connect an AI client to all of them in the same conversation. The setup uses a simple environment variable to list which project folders are registered, and optionally a bearer token for clients that support authentication. For remote access from a hosted AI service, the README recommends exposing the local server using a tunnel tool like Tailscale Funnel or a similar exposure layer. The server enforces permission policies that restrict what paths the AI can touch, along with caps on input and output sizes. An optional spawned agent mode exists for delegated work, though the README notes this is not a hard filesystem sandbox and only commands you are comfortable granting should be allowed. The project is written in TypeScript and requires Node.js 20 or newer. The README does not state a license for this project.