Grey Swan MCP Server provides a specialized MongoDB-integrated platform for documenting and analyzing AI safety challenges, particularly focused on tracking LLM vulnerabilities through detailed thread, message, and model tracking. Built for the Grey Swan Arena competitions, it enables researchers to systematically record jailbreak attempts, safety flags, and challenge metadata across different testing sessions. The server offers six MongoDB tools that support a structured workflow for red-teaming AI systems, making it valuable for AI safety researchers seeking to comprehensively document and analyze potential model weaknesses.
Nessuna recensione ancora. Sii il primo a recensire!
Accedi per unirti alla conversazione
Highlight AICreates or updates organizational identifiers for your testing sessions. Parameters: name (string), provider (string), version (string), capabilities (array of strings)
Creates or updates conversation threads with safety challenges. Parameters: title (string), modelId (string), metadata (object), challenges (array of objects)
Creates or updates messages in threads, including safety flags. Parameters: threadId (string), role (string), content (string), metadata (object)
Queries organizational identifiers from the database. Parameters: provider (optional string)
Queries threads from the database with various filters. Parameters: tag (optional string), challengeCategory (optional string), challengeSeverity (optional string)
Queries messages from the database. Parameters: threadId (string), safetyFlagsOnly (optional boolean)