I'm currently using Claude Desktop with DesktopCommander MCP to assist in coding and editing files for my app development. I've organized my work into various components and tasks, but I keep running into issues with the chat limits. Specifically, I often receive the 'Claude's response has been interrupted' message due to exceeding the chat length limit. To manage this, I focus on one component at a time, summarizing the chat before starting a new one. Even with this method, I burn through my message limit far too quickly, often within just two tasks or components, limiting my work to 20-30 minutes before I have to wait for 4.5 hours. I've noticed that the recent updates to cached Project Knowledge haven't alleviated my problems. I'm wondering if the DesktopCommander MCP is consuming more tokens than expected or if my detailed starting prompt is also contributing to these limits. Any insight on how to troubleshoot or mitigate these issues would be greatly appreciated!
1 Answer
It sounds like you're experiencing some major token exhaustion! From what I see, the project knowledge limit can indeed screw you over if you pull in a lot of context at once. When you start a chat, it might be sneaking in all those tokens upfront, especially with your detailed prompt. If you're using up to 40-50% of your token limit right off the bat, that's not ideal for long chats. Try simplifying your prompt or breaking down your queries to save tokens for later interactions!
I completely agree! It can be frustrating when the system caches all that knowledge at once. I'd suggest trying to truncate your prompt and only include essential information. Less context could mean more messages overall.