wcgw

Losing the sense of cwd is the reason why I append it in the output of each command run in wcgw mcp [1] It rarely does it incorrectly after that. I won't be surprised if claude code does the same soon. However, they do have an env flag called CLAUDE_BASH_MAINTAIN_PROJECT_WORKING_DIR=1 This should also fix the wrong dir behavior. [1] https://github.com/rusiaaman/wcgw

I believe it's not possible to restrict an LLM from executing certain commands while also allowing it to run python/bash. Even if you allow just `find` command it can execute arbitrary script. Or even 'npm' command (which is very useful). If you restrict write calls, by using seccomp for example, you lose very useful capabilities. Is there a solution other than running on sandbox environment? If yes, please let me know I'm looking for a safe read-only mode for my FOSS project [1]. I had shied away from command blacklisting due to the exact same reason as the parent post. [1] https://github.com/rusiaaman/wcgw

Awesome! I essentially started with the same premise when coding "wcgw" [1]. Give a single shell tool to chatgpt (later claude). Chronology that followed: 1. Some processes take long time: add status checking. 2. Need to kill long running process or interact with tools add option to send control characters; BashInteraction tool. 3. File editing using sed and echo is not reliable, add FileEdit tool. 4. Large output; add truncation based on tokens. 5. But wait I need command output to get truncated at the beginning but file reading `cat $file` to truncate at the end. Add special ReadFile tool. 6. File writing using 'echo "content" >file' needs unncessary quotes escaping leading to degraded quality, add special WriteFile tool. 7. Reading one file at a time is slow, change ReadFile to ReadFiles tool. 8. Directly writing files is not as good a

You can try (self promo) https://github.com/ezyang/codemcp . https://github.com/rusiaaman/wcgw is also quite popular, although they allow unrestricted shell access (that's why it's named wcgw lol).

You can basically get the same experience as aider with an MCP server like https://github.com/rusiaaman/wcgw. It's not perfect - sometimes has trouble with exact syntax of find/replace. But it's free to use up to your regular Claude subscription usage limit. I actually use it more than Cursor, because it's easier to flip back and forth between architecting/editing.

Good work. I wonder if there's really a need for separate write to terminal and read output functions? I was hoping that write command itself would execute and return the output of the command, saving back and forth latency. > and it's not always clear when the job is done I've authored a similar mcp [1] (but without terminal ui) The way I solved it is by setting a special PS1 prompt. So as soon as I get that prompt I know the task is done. I wonder if a similar thing can be done in your mcp? [1] https://github.com/rusiaaman/wcgw

For building mcp server? The official docs do a great job https://modelcontextprotocol.io/introduction My own mcp server could be an inspiration on Mac. It's based on pexpect to enable repl session and has some tricks to prevent bad commands. https://github.com/rusiaaman/wcgw However, I recommend creating one with your own customised prompts and tools for maximum benefit.

Personally I'm using the Filesystem server along with the mcp server called wcgw[0] that provides a FileEdit action. I use MacWhisper[1] to dictate. I use `tree` to give Claude a map of the directory I'm interested in editing. I usually opt to run terminal commands myself for better control though wcgw does that too. I keep the repo open in a Cursor/Windsurf window for other edits I need. But other than that I basically just tell the model what I want to do and it does it, lol. I like the Claude Desktop App interface better than trying to do things in Cursor/Windsurf directly, I like the ability to organize prompts/conversations in terms of projects and easily include context. I also honestly just have a funny feeling that the Claude web app often performs better than the API responses I get from the IDEs. [0] https://github.com/rusiaaman

It's great! I quickly reorganised my custom gpt repo to build a shell agent using MCP. https://github.com/rusiaaman/wcgw/blob/main/src/wcgw/client/... Already getting value out of it.

I have been using this custom gpt that I wrote to develop code on autopilot mode for a while now. Write code -> fix type issues -> write code on a loop. All on chatgpt.com webapp without any API access requirement so it's free. It has full shell access and can write files. I was initially hesitant to run it locally outside VM, but now I believe that if the project is version controlled, it's very hard for anything bad to happen. What could go wrong? ;)