Golf

Golf is a control layer for AI agent and MCP connections, addressing the visibility and governance gap created when teams connect tools such as Cursor, Claude Code, and Copilot to production data through MCP. It operates at the MCP layer rather than the LLM layer, governing where agents connect to data without controlling the models themselves.

Its core capabilities are discovery, enforcement, and audit. Discovery identifies AI agents, MCP servers, and data connections across an environment, including shadow infrastructure. Enforcement applies granular policies per tool, team, and data source, with controls for PII exposure, credential leaks, prompt injection, and unauthorized access, delivered at sub-millisecond latency. Audit maintains a 90-day trail of prompts, actions, and data access, pre-mapped to SOC 2, ISO 27001, NIST AI RMF, and FINRA, with evidence export. Golf integrates with identity providers via SSO, streams activity to SIEM systems, and supports on-prem, hybrid, or cloud deployment.

Golf is designed for security teams and organizations managing AI agent access to production data. It supports third-party agents including Cursor, Claude Code, GitHub Copilot, ChatGPT Enterprise, Windsurf, and custom agents, alongside any MCP server. The platform is offered through demo and assessment calls, with deployment described as taking minutes to days.