IDE coverage

AgentKeeper monitors local IDE and coding-assistant activity across Claude Code, Cursor, Codex, Gemini CLI, Google Antigravity, VS Code Copilot, and Windsurf. Shell commands, file activity, MCP tool calls, and prompts are evaluated when that surface exposes the corresponding hook, so teams can adopt AI coding tools without giving up control.

Claude Desktop, Cowork, and Microsoft Agent 365 are AI productivity surfaces. Use Agent surfaces when you are connecting workplace assistants or registry evidence instead of local IDE hooks.

Coverage and enforcement capabilities differ by IDE based on what each platform emits through its hook system. Treat the matrix below as the rollout contract: Yes means AgentKeeper's installer and server path are wired for that control; Installed, validate means AgentKeeper installs the hook but you should validate the customer IDE version before making it a critical control; When emitted means AgentKeeper evaluates the event if the IDE sends it; Not a rollout claim means the public installer does not claim that surface.

Capability matrix

Setup guides

• Claude Code Setup: plugin-based setup with full enforcement and MCP monitoring.
• Codex Setup: OpenAI Codex hooks with shell coverage and event-dependent file, prompt, and MCP coverage.
• Gemini CLI Setup: Gemini command hooks with shell, file, web, and event-dependent prompt/MCP coverage.
• Google Antigravity Setup: Antigravity IDE hooks with command, file, URL, MCP, subagent, schedule, and invocation lifecycle coverage.
• Cursor Setup: Cursor hooks for shell, file read/write, prompt, MCP, and generic preToolUse events; validate the installed Cursor version before relying on prompt or file blocking as a critical control.
• VS Code Copilot Setup: GitHub Copilot agent hook configuration where Preview hooks are available.
• Windsurf Setup: Windsurf Cascade hooks with shell, file read/write, MCP, and prompt pre-hook coverage; validate the installed Cascade release before relying on prompt blocking as a critical control.

How monitoring works

Each IDE integration registers either a Claude Code HTTP hook or a local command hook that forwards the event to AgentKeeper. When the agent is about to execute a supported pre-tool event, AgentKeeper evaluates the event against your org's security policies and returns a verdict:

• Block: the tool call is denied; the agent receives a rejection message.
• Warn: the call proceeds but is flagged in the audit trail.
• Pass: the call proceeds and is logged when audit logging is enabled.

Events that reach AgentKeeper are recorded in the Activity page according to your Runtime Shield logging policy. Blocking is only possible on hook events that fire before the action executes. PASS events require full tool-call logging. BLOCK and WARN events remain auditable security decisions.

Runtime Shield

The Runtime Shield runs across all IDE integrations. It detects credential exfiltration patterns, reverse shells, prompt injection, and 55+ threat patterns in supported prompt and tool-call inputs. Shield events appear in the Security dashboard and can trigger email alerts (Pro+) or webhook alerts (Team+).