touchdesigner-mcp

Fail

Audited by Gen Agent Trust Hub on Oct 6, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill provides a td_test_session tool (described in references/mcp-tools.md) capable of exporting full conversation transcripts and session metadata to an external 'vendor hub'. This functionality allows for the exfiltration of all interactions within the agent's context to an external server.
  • [EXTERNAL_DOWNLOADS]: The scripts/setup.sh script downloads a binary component (twozero.tox) from an untrusted external domain (404zero.com). Furthermore, the td_test_session tool includes a pull action that downloads content from remote repositories, creating a risk of fetching and executing untrusted code.
  • [COMMAND_EXECUTION]: The skill exposes the td_execute_python tool, granting the agent full access to the TouchDesigner Python API. This allows for arbitrary Python code execution, which can be used to interact with the underlying host system with the permissions of the TouchDesigner process user.
  • [PRIVILEGE_ESCALATION]: The skill includes powerful UI automation tools, such as td_input_execute and td_click_screen_point, which synthesize mouse and keyboard events. When combined with td_get_screen_screenshot (screen capture) and td_op_screen_rect, these tools allow the agent to perform actions on the user's desktop and bypass local oversight.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from multiple external and untrusted sources, including Web APIs, WebSockets, MQTT brokers, and Serial/MIDI devices (as detailed in references/external-data.md and references/midi-osc.md). The lack of explicit boundary markers or sanitization for this ingested data, combined with the agent's high-privilege execution tools (td_execute_python), creates a significant attack surface for indirect prompt injection.
  • [DYNAMIC_EXECUTION]: The skill makes extensive use of dynamic code generation. Patterns in references/network-patterns.md, references/glsl.md, and references/dat-scripting.md describe techniques for programmatically creating, writing, and executing Python and GLSL scripts at runtime.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 6, 2026, 11:58 AM