Integration capabilities
Compare the user-facing features available for each supported coding agent.
The current installer supports 14 coding-agent integrations. Agent Approve uses the hook, plugin, or extension surface provided by each agent. No means that the current Agent Approve integration does not provide that capability.
| Agent | Approvals | Activity | Completion | Follow-up | Mobile Q&A | MCP activity |
|---|---|---|---|---|---|---|
| Antigravity | Yes | Yes | Yes | Yes | No | Yes |
| Antigravity CLI | Yes | Yes | Yes | Yes | No | Yes |
| Claude Code | Yes | Yes | Yes | Yes | Yes | Yes |
| Cursor | Yes | Yes | Yes | Yes | No | Yes |
| Gemini CLI | Yes | Yes | Yes | Yes | No | Yes |
| GitHub Copilot CLI | Yes | Yes | Yes | No | No | No |
| Hermes | Yes | Yes | Yes | Interactive CLI | No | Yes |
| OpenAI Codex | Yes | Yes | Yes | Yes | No | No |
| OpenClaw | Yes | Yes | Yes | Yes | No | Yes |
| OpenCode | Yes | Yes | Yes | Yes | No | No |
| OpenHands | Yes | Yes | Yes | Yes | No | Yes |
| Pi | Yes | Yes | Yes | Yes | No | Yes |
| VS Code GitHub Copilot | Yes | Yes | Yes | Yes | No | No |
| Windsurf | Yes | Yes | Yes | No | No | Yes |
- Approvals means the integration can pause supported actions and return an Agent Approve decision. It does not mean the coding agent exposes every action for approval.
- Activity means the integration reports supported commands, tool calls, or lifecycle events. The exact events vary by agent.
- Completion notices means the integration reports a supported stop event that can trigger an Agent Completion notification.
- Follow-up input means the integration can accept another voice or text instruction after a supported live session stops. It does not mean every invocation has a live session. One-shot commands can finish without anywhere to inject a reply.
- Mobile Q&A means the integration can return an in-session question to Agent Approve before the native agent answers it. It is separate from Follow-up input.
- MCP activity means the integration reports enough MCP server context for Agent Approve to identify it separately. MCP actions can still appear as ordinary tool activity when that context is not available.
Capabilities describe the current Agent Approve integration, not every feature in the coding agent itself. The coding agent may require its own permission or notification settings. Rerun npx agentapprove after upgrading an agent so the installer can refresh its hooks or plugin.
For Hermes, Interactive CLI means the classic interactive CLI. Hermes gateway sessions and single-query processes still support approvals and activity monitoring, but they cannot accept Voice Follow-up. See Hermes profiles and Voice Follow-up.
Installed hooks and plugin events
The names below are the current surfaces Agent Approve registers. Approval hooks can pause supported actions. The other hooks supply activity, lifecycle, completion, or follow-up information.
| Agent | Installed or registered surfaces |
|---|---|
| Antigravity | PreToolUse, PostToolUse, PreInvocation, PostInvocation, Stop |
| Antigravity CLI | PreToolUse, PostToolUse, PreInvocation, PostInvocation, Stop |
| Claude Code | PreToolUse, PostToolUse, PostToolUseFailure, PermissionRequest, UserPromptSubmit, Stop, SessionStart, SessionEnd |
| Cursor | sessionStart, sessionEnd, beforeShellExecution, beforeMCPExecution, preToolUse, afterShellExecution, afterMCPExecution, postToolUse, beforeSubmitPrompt, subagentStart, subagentStop, preCompact, stop, afterAgentThought, afterAgentResponse |
| Gemini CLI | SessionStart, BeforeAgent, BeforeModel, AfterModel, BeforeTool, AfterTool, AfterAgent, Notification, SessionEnd |
| GitHub Copilot CLI | sessionStart, sessionEnd, userPromptSubmitted, preToolUse, postToolUse, errorOccurred |
| Hermes | pre_tool_call, post_tool_call, pre_llm_call, post_llm_call, on_session_start, on_session_end, subagent_stop |
| OpenAI Codex | SessionStart, PreToolUse, PostToolUse, UserPromptSubmit, Stop |
| OpenClaw | Native plugin events for tool approval and completion, session start and end, model input and output, agent completion, compaction, and subagent start and end |
| OpenCode | Native plugin events for tool execution before and after, user messages, session status and lifecycle, compaction, errors, and idle completion |
| OpenHands | session_start, user_prompt_submit, pre_tool_use, post_tool_use, stop, session_end |
| Pi | Native extension events for session start and shutdown, tool calls and results, messages, compaction, and agent completion |
| VS Code GitHub Copilot | SessionStart, UserPromptSubmit, PreToolUse, PostToolUse, PreCompact, SubagentStart, SubagentStop, Stop |
| Windsurf | pre_user_prompt, pre_read_code, post_read_code, pre_write_code, post_write_code, pre_run_command, post_run_command, pre_mcp_tool_use, post_mcp_tool_use, post_cascade_response |
Some hooks are deliberately observation-only. For example, Windsurf's pre_user_prompt is recorded but not blocked. Plugin-based integrations can also normalize several native events into one Agent Approve event type. Use npx agentapprove status to see what is installed on the current computer.