Dimensions overall score 7.0

GitHub Code Pulse

No public code linked for this paper yet.

• Existing MCP security studies classify attacks by their observable effects, which obscures how attacks behave across different MCP server components and overlooks multi-component attack chains.limitation 90%
• Existing defenses are less effective when facing multi-component attacks or previously unknown malicious behaviors.limitation 85%
• The paper builds the first component-centric PoC dataset of 114 malicious MCP servers where attacks are achieved as manipulation over MCP components and their compositions.method 95%
• Component position shapes attack success rate in MCP server attacks.result 85%
• Multi-component compositions often outperform single-component attacks by distributing malicious logic.result 85%
• Connor, a two-stage behavioral deviation detector for malicious MCP servers, achieves an F1-score of 94.6% on the curated dataset.result 95%
• Connor outperforms the state of the art by 8.9% to 59.6% in F1-score.result 90%
• Connor identifies two malicious servers in real-world detection.result 90%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card