Supported Agents

Verne runs whatever CLI agent you launch in a terminal. You install and authenticate your agents the usual way; Verne doesn’t bundle or replace them. What it adds is awareness of what each agent is doing, plus deeper integrations for the common ones.

Three levels of integration

Agents integrate at three levels, from broadest to deepest:

  1. Detection. Verne recognizes the agent running in a tab and tracks its state (working, blocked, idle) from terminal output. Every agent in the list below qualifies, and any unrecognized command still shows its foreground process.
  2. Lifecycle hooks. For agents with a hook API, Verne installs a small notify script into the agent’s config so it reports state changes precisely: on prompt submit, tool use, completion, and permission prompts. Verne sets this up and tears it down for you.
  3. MCP. Verne runs an MCP server that gives agents tools to read and write your workspace notes and to drive the embedded browser. Turn it on per agent in Settings › Agents.

Detected agents

Verne detects these agents and tracks their status:

Agent Command
Claude Code claude
Codex codex
OpenCode opencode
GitHub Copilot CLI copilot
Gemini gemini
Amp amp
Cline cline
Cursor CLI cursor
Grok grok
Hermes hermes
Kilo kilo
Kimi kimi
Kiro kiro
Antigravity antigravity
Droid droid
Pi pi
Qoder CLI qodercli

Other commands run fine; they just won’t get agent-specific status tracking.

Hooks and MCP

  • Lifecycle hooks cover the agents that support them, including Claude Code, Codex, Copilot, Gemini, and Amp. For agents without a hook API, such as OpenCode and Pi, Verne writes a small in-process plugin instead and bridges the same events.
  • MCP auto-install is available in Settings › Agents for Claude Code, Codex, Cursor CLI, and OpenCode. For any other agent, copy the install commands and register the server by hand.

Next steps