You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 45380a6
Browse filesBrowse the repository at this point in the historyBrowse files
Copy file name to clipboardExpand all lines: README.md
+6Lines changed: 6 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -396,6 +396,12 @@ mcpm baseline remains 0.47.1. It exercises
396
396
actual global configurations and MCP calls with synthetic inputs; live models,
397
397
interactive chat and other client/platform combinations require separate tests.
398
398
399
+
v0.47.3 fixes VS Code user-global JSONC configuration edits so unrelated
400
+
comments/settings and the raw backup survive the MCP lifecycle. It also adds
401
+
separate native approval/denial tests: four cases each in the pinned VS Code
402
+
and Claude Code builds, with synthetic model responses. Native macOS dialogs,
403
+
workspace/server trust, inline chat/Agent Host and live providers remain unverified.
404
+
399
405
Install-time trust scoring catches most poisoned servers before they ship. But what about **rug-pulls** — a server that changes its tool definitions after you've already approved them? Or **prompt-injection in tool responses** — adversarial text embedded in a Slack message, web page, or calendar invite that the agent reads through your trusted MCP server?
400
406
401
407
`mcpm guard`adds a runtime inspection layer. It wraps every installed MCP server with a stdio relay, scans tool descriptions / responses / arguments for OWASP MCP Top 10 attack patterns, pins each tool's schema at install time, and blocks calls when the live response drifts from the pin (rug-pull defense).
0 commit comments