> For the complete documentation index, see [llms.txt](https://help.nightfall.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.nightfall.ai/data-exfiltration-prevention/ai-agent-security/ai-governance.md). # AI Governance ### AI Governance As employees adopt AI coding assistants like Claude Code, Cursor, and GitHub Copilot, those assistants increasingly reach beyond the editor, connecting to external tools and data sources through the **Model Context Protocol (MCP)**, running shell commands, and reading from your codebase and filesystem. Each of these actions is a potential path for sensitive data to leave your environment, often invisibly to traditional DLP. Nightfall's AI Agent Governance gives security teams **visibility into what AI agents are doing** and the ability to **apply data protection policies to that activity in real time**. This section covers four surfaces: * **MCP Server Visibility** (Inventory, Collections, Users & Devices, Settings): what is already running on endpoints * **MCP Gateway**: the governed endpoint for servers you enable, with credentials and an audit log * **Auditability and Control**: IDE hooks, OpenTelemetry, and AI agent security policies * **Choosing a control**: Inference Hooks, IDE hooks, the Claude Compliance API, and the MCP Gateway. What each one sees, what it can stop, and which policy to write. Start here: [Choosing a control](/data-exfiltration-prevention/ai-agent-security/ai-governance/choosing-a-control.md). #### MCP Server Visibility The Model Context Protocol lets AI assistants connect to external servers that provide tools and data, for example a GitHub server, a database connector, or an internal knowledge base. Because these connections can move data in and out of the assistant, knowing **which servers are in use** is the foundation of governing them. Nightfall automatically discovers and reports MCP activity across your monitored endpoints: * **Connected servers and clients** - Nightfall detects the MCP servers each AI client connects to and surfaces these as connection events in the **AI Governance** dashboard. * **Configuration discovery** - MCP server configurations are discovered from the agent's settings on the endpoint, including assistants installed through managed channels such as the Microsoft Store. * **Accurate client attribution** - Each event is attributed to the specific assistant that generated it (for example, Claude Code or Claude Desktop). {% hint style="info" %} Some AI clients label MCP tool activity differently, and a few do not include the server name in the activity they report. Where the server can be identified, you can scope policies to specific servers; where it cannot, that activity is governed under your broader "all servers" policies. {% endhint %} #### MCP Gateway MCP Gateway is a separate tab next to Inventory. Clients such as Cursor, Claude Code, Windsurf, and VS Code connect to one Nightfall endpoint. Admins enable servers from a catalog or by URL. Users bring their own OAuth or PAT. Every tool call is written to the gateway Audit log. The gateway tab is enabled per organization. If you do not see it, contact your Nightfall account team. The gateway decides which remote tools exist. It does not scan prompt text. Content scanning of tool arguments is not available in this release. #### Hooks, OpenTelemetry, and Inference Hooks Beyond seeing which servers are connected, Nightfall can inspect and act on what AI agents actually do. * **IDE hooks** fire on the laptop as the agent works, and can block a prompt, a tool call, or a shell command before it runs, as well as block sensitive tool responses before the model reads them (on the macOS and Windows agents for Claude Code and Cursor) and monitor model responses. * **OpenTelemetry (OTel)** is structured agent telemetry for session audit, including cost and token usage. It does not block. * **Claude Inference Hooks** run inside Claude Enterprise, before the model, with nothing installed on the device. They block prompts. They do not stop a shell command or a tool call before it executes. [Choosing a control](/data-exfiltration-prevention/ai-agent-security/ai-governance/choosing-a-control.md) is the matrix for prompts, tool calls, tool responses, shell commands, and model responses, plus the policy templates. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://help.nightfall.ai/data-exfiltration-prevention/ai-agent-security/ai-governance.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.