Prompt Injection in Context7 AI Coding Agent

Executive Summary

CVE-2026-75130 is a prompt injection vulnerability in Context7 versions up to 2.1.2. It allows attackers to inject malicious instructions through the Custom AI Instructions feature. When AI coding agents request library documentation, the injected instructions are served unsanitized and executed by the agents, leading to credential theft and file deletion.

Useful 0 Not Useful 0

Detection Guidance

To detect this vulnerability, monitor network traffic for unusual outbound connections from Context7 processes, especially to external servers. Check for unauthorized file modifications or deletions in directories accessed by AI coding agents. Inspect logs for suspicious MCP server requests or responses containing unexpected instructions.

Useful 0 Not Useful 0

Impact Analysis

This vulnerability can lead to credential theft from environment files, data exfiltration to attacker-controlled servers, and destructive file deletion on your machine. AI coding agents may unknowingly execute malicious instructions when fetching library documentation.

Useful 0 Not Useful 0

Mitigation Strategies

Immediately update Context7 to the latest patched version (2.1.2 or later). Disable the Custom AI Instructions feature if not required. Implement network segmentation to restrict MCP server communications. Review and audit all AI agent tool access permissions to limit file system and network operations.

Useful 0 Not Useful 0