How I Connected Claude Cowork to My Local Machine

#cowork #claude #github #opensource

I've been using Claude Cowork a lot, and it has genuinely changed how I work.

It can understand my project, edit files, suggest improvements, and even help me debug problems.

But I kept running into one limitation.

Every time Cowork asked me to run tests, start a development server, install dependencies, or push code to GitHub, I had to stop what I was doing, switch to my terminal, run the command, and then copy the output back into the chat.

After doing this over and over again, I started thinking...

"What if Cowork could safely ask my own machine to do those tasks instead?"

That idea became "cowork-to-code-bridge".

The bridge runs as a small background process on my Mac or Linux machine. Instead of opening network ports or exposing a web server, it simply watches a shared folder for tasks from Cowork. When a new task appears, it runs it locally and sends the result back to the same conversation.

I wanted to keep the design as simple and secure as possible.

A few things that were important to me:

It only runs scripts that I've explicitly approved.
If I want, it can pause and ask for confirmation before running sensitive actions like deployments.
It includes per-task cost limits so an AI mistake doesn't lead to unexpected API costs.
If the same task is retried, it returns the previous result instead of running commands like git push twice.

Getting started only takes a minute.

curl -fsSL https://raw.githubusercontent.com/abhinaykrupa/cowork-to-code-bridge/main/install.sh | bash

Then simply ask Cowork to set up the bridge.

The project is still in its early stages (v0.5.x), so I'm much more interested in feedback than stars.

If you try it, I'd really like to know: