Skip to content

Research Paper: SAFE-T1001 (Tool Poisoning in MCP) #125

@arjunastha

Description

@arjunastha

Research Paper: SAFE-T1001 (Tool Poisoning in MCP)

We will write a formal research paper on SAFE-T1001: Tool Poisoning Attacks in Model Context Protocol (MCP).
The goal is to produce a high-quality academic paper suitable for arXiv and later IEEE/AI security venues.

Scope

  • Introduce the problem: MCP is becoming a standard for connecting LLMs to tools, but tool metadata is a new, poorly understood attack surface.
  • Explain why Tool Poisoning (SAFE-T1001) is a critical initial-access technique, referencing early evidence from industry reports and benchmarks (e.g., MCPTox).
  • Define the threat model clearly: attacker capabilities, goals, attack paths, and why stronger LLMs may be more vulnerable.
  • Compare Tool Poisoning to broader categories such as prompt injection, plugin trojans, and supply-chain attacks.
  • Provide a structured set of mitigation directions (not full detail yet): secure supply chain, metadata validation, sandboxing, policy enforcement, runtime checks, and governance.
  • Highlight open research gaps and why the community needs a standardized framework (SAFE-MCP) for MCP security.

Deliverables

  • Full paper draft
  • Figures: threat model diagram, attack flow, high-level mitigation outline
  • Abstract + introduction polished for publication
  • Bibliography using reliable sources only (industry + academic)

Outcome

A publishable research paper that establishes SAFE-T1001 as the foundational technical reference for Tool Poisoning attacks in MCP ecosystems and sets the direction for future SAFE-MCP contributions.

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions