fff sometimes dies

[charger89]

Noticed that claude still tends to use grep over fff quite a bit. Asked the model about it:

in the bash commands you run, I can see still many references to grep command, whereas the ~/.config/claude/CLAUDE.md file states: "For any file search or grep in the current git-indexed directory, use fff tools exposed via MCP.". Why is fff MCP not used instead of find/grep and /snip/

And the answer was:

Two separate reasons, depending on which specific commands you're looking at:

1. The fff MCP server has been disconnecting and reconnecting during this session. There were explicit system notices that mcp__fff__find_files, mcp__fff__grep, and mcp__fff__multi_grep became unavailable ("their MCP server disconnected... do not search for them"), and later a notice that it reconnected. During the windows where it was disconnected, those tools genuinely weren't callable, so grep/find via Bash were the only option.

2. Much of the searching wasn't actually in the git-indexed project directory, which is what the CLAUDE.md rule scopes to. A large share of the grep/find/unzip commands were run against extracted dependency library sources sitting in the scratchpad temp directory (/tmp/.../scratchpad/kc-sources) and Gradle's dependency cache (~/.config/gradle/caches/...) — decompiled/extracted third-party jar sources, not part of this repo at all. fff indexes the project's own git repo; it wouldn't have anything useful to say about files that live outside it, so reaching for grep/find/unzip there isn't a violation of the rule, it's just outside its scope.

---

Sounds valid. My subsequent questions are:
a) why would the fff be disconnection per the 1st answer above? How to debug/fix it?
b) is there a good reason why the readme-given instruction tells to scope fff-usage to "git-indexed directory"? Is there any downside in using it globally, everywhere?

[dmtrKovalenko]

This is now in work and will be fixed in the next release