Hacking Your Own AI Coding Assistant with Claude Pro and MCP - https://news.ycombinator.com/item?id=43410866

I started playing around with it yesterday evening. Running Claude Desktop under manjaro linux with some tweaks, I did managed to activate codemcp.

As I am also quite new with mcp and how to proper use the protocol in tools, I am not sure I did everything correct. I did also install filesystem, git, github and sequential-thinking along side.

However, git will not work. The addition of the mcp git server might not be needed? But without it, it did also not work

However, what it has done so far is mind blowing. I did create a long query the other day about a project idea. This generated a road map, detailed implementation instruction, guidlines, all that is needed to complete the problem. I let it Claude Desktop read in the files with help of codemcp.

Claude Dekstop have been working for multiple hours before I seem to have hot some kind of limit. I now get the message to type continue, but it always start the with the same task.

And what it has been created is awesome. The frontend is nearly finished, but it lacked to create the startup code first. Overall I am blow away what I can do with mcp and esp. with this small lib the author has created here. Code quality is good and event though I had multiple limits, where I needed to type continue, the code is logically make sense. Really good!

I also do not yet know how I can continue a old session.

P.S. @anthropic, If you are reading this. I really do not want to abuse your service. Please, instead of giving me no answer, or beeing stuck in an infinite loop, give me a proper return code and a error, that I reached my limit for the next X minutes or so. I would really love this

If you need to start a new chat, it works just fine. Tell Claude what's happened so far and what you want it to do. You can also ask Claude to summarize the old conversation, that's how /compact in claude code works too.