A bug in Claude Code's MCP (Meta-Command Protocol) system, fixed in version 2.1.299, allowed certain allow rules to inadvertently grant access to unintended servers. Specifically, an allow rule for 'mcp__a' could also match tools from a server named 'a__b' or 'a_', leading to potential security risks as these connections bypassed user permission prompts. The fix ensures that MCP allow rules now precisely target the named server, preventing such naming collisions and enhancing security by requiring explicit permission for each server connection. AI
IMPACT Addresses a specific security vulnerability in an AI coding assistant's tool interaction mechanism, enhancing the safety of its agent framework.
RANK_REASON This item details a specific bug fix and its implications for a particular feature (MCP) within a software product (Claude Code), rather than a major release or industry-wide event.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →