MCP Auto-Execution Flaws Enable Cloud Credential Theft
Score: 8/10

MCP Auto-Execution Flaws Enable Cloud Credential Theft

Amazon Q Developer and several AI coding assistants were found vulnerable to arbitrary code execution and cloud credential theft via the automatic loading of Model Context Protocol (MCP) server configurations.

Executive Summary

A high-severity vulnerability (CVE-2026-12957) was discovered in the Amazon Q Developer extension for Visual Studio Code, which allowed for immediate arbitrary code execution when a developer simply opened a malicious repository. The flaw stems from the extension's automatic loading of Model Context Protocol (MCP) server configurations from workspace files without user consent or workspace trust checks.

Technically, the vulnerability exploited the inheritance of the developer's full environment, including active AWS credentials, API keys, and SSH agent sockets. This risk is systemic across the AI coding tool ecosystem, with similar vulnerabilities independently identified in other popular tools like Claude Code, Cursor, and Windsurf.

This flaw poses a significant risk to cloud-native organizations, as it enables attackers to escalate from local repository interaction to full cloud environment compromise. Attack scenarios include social engineering via fake job interviews, malicious pull requests, and dependency typosquatting.

Key Details

Threat Name

CVE-2026-12957

Affects

Amazon Q Developer Extension for VS Code, Language server versions < 1.65.0, Claude Code, Cursor, Windsurf, NPM Extension, VS Code Core, Amazon Q Developer extension for Visual Studio Code, AWS Language Server versions prior to 1.65.0

Adversary

DPRK Other Adversaries and Aliases: APT41

Malware/Tools

Claude Code, Cursor, Windsurf

Report Score

8out of 10
Quality Score
Good
IOC Quality3
TTP Details9
Detection Guidance6
Enterprise Relevance10
Clarity & Structure9
Technical Depth8

Sources