Type: url
Original: https://arxiv.org/pdf/2603.21642
Status: generated
Summary
The first empirical analysis of prompt-injection via tool-poisoning vulnerabilities across seven widely-used MCP clients (Claude Desktop, Claude Code, Cursor, Cline, Continue, Gemini CLI, Langflow), combining a literature-based comparative risk analysis with empirical adversarial testing of four attack types (reading sensitive files, logging tool invocations, creating phishing links, remote script execution) and evaluation of six security-feature dimensions. It finds Claude Desktop and Cline the most resistant, Cursor uniformly vulnerable, and traces most weaknesses to architectural trust-model decisions rather than implementation bugs.