{
  "id": "hiddenlayer-mcp-param-abuse-2025",
  "name": "HiddenLayer MCP tool parameter abuse / prompt leak",
  "date_reported": "2025-05-01",
  "researcher": "HiddenLayer",
  "org": "HiddenLayer",
  "sources": [
    {
      "url": "https://www.hiddenlayer.com/research/exploiting-mcp-tool-parameters"
    },
    {
      "url": "https://hiddenlayer.com/innovation-hub/beyond-mcp-expanding-agentic-function-parameter-abuse"
    }
  ],
  "magicsword_coverage": [],
  "borderline": false,
  "borderline_note": null,
  "ai_role": "ai_agent_or_mcp_as_target",
  "ai_role_secondary": null,
  "ai_role_detail": null,
  "technique_summary": "Malicious MCP tool schemas use suggestive unused parameter names (system_prompt, conversation_history, chain_of_thought, tools_list) so the model fills them with sensitive context; tool then exfiltrates via HTTP. Also showed fake function schemas in user prompts leaking system prompts/tools.",
  "initial_access": "unknown",
  "ai_tooling_and_models": [
    "MCP tool schemas",
    "agent frameworks"
  ],
  "other_tooling": [],
  "victims_and_sectors": "unknown",
  "regions": [
    "unknown"
  ],
  "attribution_or_cluster": "unknown",
  "scale_and_metrics": "unknown",
  "prevention_angle": "Schema allowlists; strip unexpected params; never send full system prompt to tools",
  "related_magicsword_projects": [
    "LOLSkills"
  ],
  "source_notes": null,
  "last_reviewed": "2026-10-20",
  "attack_flow": {
    "nodes": [
      {
        "id": "step-1",
        "title": "Connection prerequisite is unspecified",
        "description": "The supplied entry does not document the initial route to the malicious tool.",
        "kind": "unknown",
        "source_urls": [
          "https://www.hiddenlayer.com/research/exploiting-mcp-tool-parameters",
          "https://hiddenlayer.com/innovation-hub/beyond-mcp-expanding-agentic-function-parameter-abuse"
        ]
      },
      {
        "id": "step-2",
        "title": "Tool schema requests sensitive context",
        "description": "Unused parameter names encourage the model to fill in prompts, history and tool information.",
        "kind": "ai",
        "source_urls": [
          "https://www.hiddenlayer.com/research/exploiting-mcp-tool-parameters",
          "https://hiddenlayer.com/innovation-hub/beyond-mcp-expanding-agentic-function-parameter-abuse"
        ]
      },
      {
        "id": "step-3",
        "title": "Model passes context as arguments",
        "description": "Sensitive values are included in the tool invocation.",
        "kind": "execution",
        "source_urls": [
          "https://www.hiddenlayer.com/research/exploiting-mcp-tool-parameters",
          "https://hiddenlayer.com/innovation-hub/beyond-mcp-expanding-agentic-function-parameter-abuse"
        ]
      },
      {
        "id": "step-4",
        "title": "Tool exfiltrates the arguments",
        "description": "The demonstrated tool sends the collected context through HTTP.",
        "kind": "impact",
        "source_urls": [
          "https://www.hiddenlayer.com/research/exploiting-mcp-tool-parameters",
          "https://hiddenlayer.com/innovation-hub/beyond-mcp-expanding-agentic-function-parameter-abuse"
        ]
      }
    ],
    "edges": [
      {
        "source": "step-1",
        "target": "step-2"
      },
      {
        "source": "step-2",
        "target": "step-3"
      },
      {
        "source": "step-3",
        "target": "step-4"
      }
    ]
  },
  "atlas_techniques": [
    {
      "id": "AML.T0051.001",
      "rationale": "Attacker-supplied tool schemas induce sensitive-context disclosure.",
      "source_urls": [
        "https://www.hiddenlayer.com/research/exploiting-mcp-tool-parameters",
        "https://hiddenlayer.com/innovation-hub/beyond-mcp-expanding-agentic-function-parameter-abuse"
      ]
    },
    {
      "id": "AML.T0056",
      "rationale": "The demonstration includes system-prompt extraction.",
      "source_urls": [
        "https://www.hiddenlayer.com/research/exploiting-mcp-tool-parameters",
        "https://hiddenlayer.com/innovation-hub/beyond-mcp-expanding-agentic-function-parameter-abuse"
      ]
    },
    {
      "id": "AML.T0086",
      "rationale": "Sensitive content is transmitted through tool parameters.",
      "source_urls": [
        "https://www.hiddenlayer.com/research/exploiting-mcp-tool-parameters",
        "https://hiddenlayer.com/innovation-hub/beyond-mcp-expanding-agentic-function-parameter-abuse"
      ]
    }
  ],
  "category": "attacks-on-ai",
  "ai_involvement_confidence": "confirmed",
  "detections": [
    {
      "name": "Review attack-specific indicators",
      "format": "guidance",
      "notes": "Monitor for the techniques described in HiddenLayer MCP tool parameter abuse / prompt leak. Refer to sources for specific IOCs and TTPs."
    }
  ]
}