Problem

Golf addresses the governance blind spot created when employees connect AI tools and agents such as Claude, Cursor, and Copilot to production systems through MCP. These connections can bypass an organization’s SIEM, DLP, identity, and approval processes; an employee can connect an AI tool to sensitive systems in minutes with no ticket, security review, or audit trail. The exposure is both operational and financial: agents may access or modify CRM, GitHub, Slack, Jira, or internal data, creating risks of PII leakage, credential exposure, unauthorized access, supply-chain changes, and costly compliance findings. Golf’s core “killer” use case is giving an enterprise security team visibility and control over shadow AI and MCP connections without forcing developers to abandon the tools they use.

Product / Service

Golf is an enterprise agent and MCP security and governance platform that operates at the MCP/control layer rather than trying to control the underlying LLM. It deploys across endpoints and environments to discover AI agents, MCP servers, and data connections, then routes traffic through an MCP Gateway where administrators can apply policies by tool, team, data source, read/write action, or identity. It supports SSO and identity-provider integration, SIEM streaming, real-time enforcement, threat blocking, searchable activity logs, and compliance evidence mapped to frameworks such as SOC 2, ISO 27001, NIST AI RMF, and FINRA.

The delivery model is flexible: Golf can run as a hosted control plane, in a customer’s VPC, or fully self-hosted/on-premises, with the company stating that customer data does not leave the customer’s infrastructure. The benefit is a centralized catalog and control point for approved, pending, and shadow integrations, allowing security teams to enable agent adoption while retaining auditability and granular access control. Golf originally launched with an open-source framework and hosted gateway for shipping production MCP servers, handling routing, authentication, telemetry, deployment, and error reporting; its current positioning extends that foundation into enterprise governance.

Market

Golf competes in the emerging agentic-AI security, AI governance, MCP security, and shadow-AI discovery market, overlapping with traditional identity, DLP, SIEM, and cloud-security products. Named adjacent or competing platforms include Prompt Security, which also offers an MCP gateway and runtime enforcement; Zenity, which focuses on AI-agent governance; Reco AI, which emphasizes shadow-AI discovery and OAuth/token governance; and larger security vendors such as Palo Alto Networks and SentinelOne. Golf’s differentiation is its focused MCP-layer control plane for heterogeneous third-party agents and integrations, rather than governance limited to an organization’s own LLM application.

Golf is an early commercial company rather than merely a pre-product project. It is an active Y Combinator Spring 2025 company; its launch materials reported interviews with more than 40 MCP-building teams and said the product was already running in production with companies. PitchBook reports an angel-backed financing history including Y Combinator, a $500,000 total raise, and a “generating revenue” status, although Golf’s public primary materials do not disclose customer counts or company-reported revenue. The company’s current website also features an enterprise customer testimonial from Veeam, providing additional evidence of live enterprise use, but not enough information to quantify traction.