Golf
Enterprise MCP Control Plane
746 followers
Enterprise MCP Control Plane
746 followers
Golf is the enterprise control plane for MCP. It gives security and IT teams full visibility into how AI connects to enterprise systems — with policy enforcement, real-time threat blocking, and a complete audit trail. Discover, enforce, audit. End-to-end.
This is the 2nd launch from Golf. View more
Golf
Launched this week
Govern and secure AI agents and MCP servers with centralized visibility, policy control, and audit trails. Security, compliance, and control for the agentic era.
Payment Required
Launch Team / Built With
This is the layer everyone is going to need and nobody is building yet. The stat about finding 150 MCP servers where 50 had destructive production access and nobody on the security team knew they existed is wild but also exactly what I'd expect.
Gianmarco's question about multi-agent audit trail attribution is the hard one. When an orchestrator spawns sub-agents that each hit MCP tools, tracking causality through that chain is brutal. The governance layer basically has to reconstruct the agent's decision tree in real time, not just log individual tool calls. Are you doing that already, or is attribution still at the session level right now?
MCP governance is going to be table stakes in 6 months. Good timing on this.
Golf
bunny.net
@wbbw1 awesome work, congrats on the launch!
Golf
@marek_nalikowski thanks a lot!
@wbbw1 congrats!
The "no end-to-end governance layer" observation is exactly right — most enterprise MCP security today is point solutions bolted on. The question that gets interesting at scale: how does Golf handle agent identity in multi-agent chains? If an orchestrator spawns five sub-agents that each call MCP tools, does the audit trail attribute actions to the orchestrator, each sub-agent individually, or the human session that triggered the chain? That attribution layer seems like the hardest part to get right — and the one that makes the difference between a compliance checkbox and a tool a SOC team actually trusts.
Golf
@giammbo Great question! Golf governs employees connecting your internal systems to third-party AI tools via MCP. Think an engineer using Cursor or Claude Desktop hitting your internal Notion, GitHub, or production database through MCP. In that case, you don't control anything - neither the agent, nor the tools your employees are connecting to them.
In that model, attribution is actually clean: every MCP tool call is tied to a real employee identity through your IDP. The audit trail shows you which person, using which agent, called which tool, on which system — with the full request and response. When compliance asks, "who let Claude touch customer data last Tuesday?" - you have a name, a timestamp, and the exact action.
@antoni_gmitruk1 That makes attribution much cleaner in the single-user case. The edge I keep thinking about: as Cursor and Claude Desktop get more agentic, an engineer might kick off a task and step away while the agent runs 30 tool calls autonomously. Same IDP identity — but very different oversight level. Does Golf capture anything at the session level that flags autonomous execution vs direct user-initiated calls?
Golf
@vouchy Yeah, honestly, not sure who was more surprised, us or them. The scanner doesn't lie though 😅
Told
The 'control plane' framing is smart — as MCP adoption scales across enterprise teams, the governance layer is often an afterthought until something goes wrong. Centralized audit trails for agentic actions in particular will be a real selling point for compliance-heavy orgs. Curious how you handle policy conflicts when multiple teams have deployed agents with overlapping permissions — is that a manual resolution process or something Golf enforces automatically? Also wondering if the tooling surfaces enough context in the audit trail for non-technical stakeholders (legal, infosec) to actually act on what they're seeing.
Trufflow
Visibility is so important. People forget things all the time. Having to deal with "orphaned" MCPs that could become the next security risk is definitely not ideal.
Most enterprise infra tools I've tracked struggle with the validation paradox: enterprises want proven scale, but scale requires enterprise adoption first. Are you seeing traction through bottom-up developer adoption or top-down enterprise sales?