Hi @olaservo (apologies for the direct ping — picking you because you merged the recent filesystem fixes #3229 and #3434),
Filing as a process question rather than a bug. Issue #4157 (edit_file newText interpreted as String.prototype.replace replacement-pattern) currently has 8 open PRs targeting the same root cause:
All target the same JS gotcha — passing a string as the replacement makes String.prototype.replace interpret $1 / $& / $' / $` / $$ as substitution patterns. Most pick one of two approaches: (a) escape $ → $$ before passing, or (b) pass a callback () => normalizedNew which is interpreted verbatim per the MDN spec. The 5 PRs at +17/-1 are largely the callback-form variant (smallest diff).
Wanted to surface this as a queue-depth signal — contributors are clearly motivated to fix this (probably hitting it in their own MCP server work), but each new PR fragments review attention further. Happy to help close the duplicate set as won't merge once you pick a canonical one, if that would unblock you.
(Context: I'm not the author of any of these, just a contributor who's been working through the MCP SDK queues and noticed the pile-up.)
Hi @olaservo (apologies for the direct ping — picking you because you merged the recent filesystem fixes #3229 and #3434),
Filing as a process question rather than a bug. Issue #4157 (
edit_filenewText interpreted asString.prototype.replacereplacement-pattern) currently has 8 open PRs targeting the same root cause:All target the same JS gotcha — passing a string as the replacement makes
String.prototype.replaceinterpret$1/$&/$'/$`/$$as substitution patterns. Most pick one of two approaches: (a) escape$→$$before passing, or (b) pass a callback() => normalizedNewwhich is interpreted verbatim per the MDN spec. The 5 PRs at +17/-1 are largely the callback-form variant (smallest diff).Wanted to surface this as a queue-depth signal — contributors are clearly motivated to fix this (probably hitting it in their own MCP server work), but each new PR fragments review attention further. Happy to help close the duplicate set as
won't mergeonce you pick a canonical one, if that would unblock you.(Context: I'm not the author of any of these, just a contributor who's been working through the MCP SDK queues and noticed the pile-up.)