mcp-servers-for-revit-integration

Warn

Audited by Gen Agent Trust Hub on Sep 16, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The send_code_to_revit tool allows for the runtime execution of arbitrary C# code within the Revit application process. This capability allows an agent to compile and run any logic provided in the code argument.
  • [REMOTE_CODE_EXECUTION]: The installation instructions promote the use of npx -y mcp-server-for-revit, which downloads and executes code from the public npm registry. Additionally, the send_code_to_revit functionality serves as a potential sink for remote code execution if the agent's instructions are compromised.
  • [EXTERNAL_DOWNLOADS]: The skill directs users to download and install binary ZIP files from a GitHub repository (mcp-servers-for-revit/mcp-servers-for-revit) and uses unpinned npm packages for server setup.
  • [DATA_EXFILTRATION]: The export_room_data tool allows the assistant to write Revit model data to a user-defined outputPath. If misused, this can be used to write sensitive information to unsafe locations or directories monitored by external services.
  • [COMMAND_EXECUTION]: The query_stored_data tool accepts raw SQL query strings, which could be exploited for unauthorized database access or modification if input is not strictly validated.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes Revit model metadata (element names, parameters, and comments) which could contain adversarial instructions intended to influence the agent when it queries the model.
  • Ingestion points: Tools such as get_current_view_elements, get_selected_elements, and ai_element_filter retrieve strings and properties directly from Revit elements.
  • Boundary markers: No delimiters or explicit instructions to ignore embedded commands are documented to protect the data/instruction boundary.
  • Capability inventory: The skill has the ability to execute C# code (send_code_to_revit), delete elements (delete_element), and perform file writes (export_room_data).
  • Sanitization: No evidence of sanitization or character escaping is provided for the data retrieved from Revit before being presented to the AI context.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 16, 2026, 10:43 PM
Security Audit — agent-trust-hub — mcp-servers-for-revit-integration