fix(mcp): prevent byte-slicing multi-byte UTF-8 in excerpt()

[hannanmax]

Problem

excerpt() uses len(s) (byte count) for the length check and s[:n] (byte slice) for truncation. For non-ASCII text — emoji, RTF markup, or non-Latin server descriptions — this splits mid-rune, producing an invalid UTF-8 string that JSON-encodes to garbage or a Unicode replacement character.

Minimal repro:

excerpt("héllo", 3)  // returns "h\xc3" — invalid UTF-8, not "hél"

Fix

Convert to []rune so both the length check and the slice operate on Unicode codepoints:

func excerpt(s string, n int) string {
    r := []rune(s)
    if len(r) <= n {
        return s
    }
    return string(r[:n]) + "…"
}

[hannanmax]

Hey! Loved the repo - really clean approach to black-box agent recon with zero deps. Found a few bugs while reading through the code and raised 5 PRs (#4, #5, #6, #7, #8). Happy to discuss any of them if you have questions!

[hackwither]

hey, thanks for the contributions! :D