# Coding agents side by side

URL: https://agentvera.dev/en/agents/comparison/

What each of the 18 coding agents supports inside AgentVera, in one table. It is built from the agent pages; pick an agent for the details.

| Agent | Made by | Command | Effort levels | MCP and browser | Account chains | Copy conversation | /compact |
| --- | --- | --- | --- | --- | --- | --- | --- |
| [Claude Code](https://agentvera.dev/en/agents/claude-code/) | Anthropic | `claude` | low, medium, high, xhigh, max | Yes | Yes | Yes | Yes |
| [Codex](https://agentvera.dev/en/agents/codex/) | OpenAI | `codex` | low, medium, high, xhigh | Yes | Yes | Yes | Yes |
| [Gemini CLI](https://agentvera.dev/en/agents/gemini-cli/) | Google | `gemini` | None | Yes | No | No | No |
| [Copilot](https://agentvera.dev/en/agents/copilot-cli/) | GitHub | `copilot` | minimal, low, medium, high, xhigh, max | Yes | No | No | Yes |
| [Cursor](https://agentvera.dev/en/agents/cursor-agent/) | Cursor Agent | `cursor-agent` | None | Yes | No | No | No |
| [Antigravity](https://agentvera.dev/en/agents/antigravity/) | Google | `agy` | low, medium, high | macOS, Linux | No | No | No |
| [opencode](https://agentvera.dev/en/agents/opencode/) | Any provider | `opencode` | None | Yes | No | Yes | Yes |
| [Hermes](https://agentvera.dev/en/agents/hermes-agent/) | Nous Research | `hermes` | None | Yes | No | No | No |
| [Qwen Code](https://agentvera.dev/en/agents/qwen-code/) | Alibaba | `qwen` | None | Yes | No | No | No |
| [Kimi Code](https://agentvera.dev/en/agents/kimi-code/) | Moonshot AI | `kimi` | low, medium, high, max | Yes | No | No | No |
| [Kiro](https://agentvera.dev/en/agents/kiro-cli/) | AWS | `kiro-cli` | low, medium, high, xhigh, max | Yes | No | No | No |
| [Amp](https://agentvera.dev/en/agents/amp/) | Sourcegraph | `amp` | low, medium, high, ultra | Yes | No | No | No |
| [Droid](https://agentvera.dev/en/agents/factory-droid/) | Factory | `droid` | low, medium, high, xhigh, max | Yes | No | No | No |
| [Goose](https://agentvera.dev/en/agents/goose/) | Block | `goose` | low, medium, high, max | Yes | No | No | No |
| [Vibe](https://agentvera.dev/en/agents/mistral-vibe/) | Mistral | `vibe` | None | Yes | No | No | No |
| [Crush](https://agentvera.dev/en/agents/crush/) | Charm | `crush` | None | Yes | No | No | No |
| [Aider](https://agentvera.dev/en/agents/aider/) | Open source | `aider` | low, medium, high | No | No | No | No |
| [Cline](https://agentvera.dev/en/agents/cline/) | Cline CLI | `cline` | low, medium, high, xhigh | Yes | No | No | No |

- Effort levels: the reasoning levels you can pick for the agent in AgentVera. “None” means the CLI has no effort setting or it follows the model you pick.
- MCP and browser: the MCP servers registered in AgentVera and AgentVera’s browser are given to the agent on every launch.
- Account chains: moving to the next account when a limit is hit, and context profiles.
- Copy conversation: copy the conversation into a new agent and carry on there.

## Agent by agent

**Claude Code**

- Sign-in: The default account uses your existing Claude Code login. Every account you add gets its own CLAUDE_CONFIG_DIR folder; you sign in with claude auth login in the app’s terminal.
- Model: Models are listed by version from Claude Code’s local model catalog. Fast mode can be turned on for models that support it.
- Effort: Auto, low, medium, high, extra high and max, picked with a slider.
- Session resume: Each agent starts with its own session id (--session-id) and continues with --resume. The conversation can be copied into a new agent with --fork-session.
- MCP: The MCP servers registered in AgentVera and the AgentVera browser (Playwright MCP) attach on every launch. Claude in Chrome is also available.

**Codex**

- Sign-in: The default account uses your existing Codex login. Every account you add gets its own CODEX_HOME folder; you sign in with codex login in the app’s terminal.
- Model: Models are read live with codex debug models.
- Effort: The effort levels the selected model offers, picked with a slider.
- Session resume: Each agent gets a stable conversation key; when the app reopens, the agent continues from the CLI’s local history. A conversation can be forked into a new agent with codex fork.
- MCP: The MCP servers registered in AgentVera and the AgentVera browser attach on every launch.

**Gemini CLI**

- Sign-in: Gemini CLI uses its own Google login or GEMINI_API_KEY; the agent’s terminal asks which on first launch.
- Model: Left empty, Gemini CLI’s default model is used; a chosen model is passed with --model.
- Effort: Gemini CLI has no effort levels.
- Session resume: Each agent starts with a session id from AgentVera (--session-id); a saved session opens with --resume.
- MCP: AgentVera’s MCP servers and browser attach on every launch, merged by name with your own Gemini settings.

**Copilot**

- Sign-in: Copilot uses its own GitHub login; if you’re signed out, run /login in the agent’s terminal.
- Model: Left empty, Copilot’s default model is used; a chosen model is passed with --model.
- Effort: Minimal, low, medium, high, extra high and max (--reasoning-effort).
- Session resume: Each agent starts with a session id from AgentVera (--session-id); a saved session opens with --resume.
- MCP: AgentVera’s MCP servers and browser are added on every launch with --additional-mcp-config.

**Cursor**

- Sign-in: Cursor uses its own CLI session; if you’re signed out, the agent’s terminal asks you to sign in on first launch.
- Model: Left empty, the model from your Cursor settings is used. List models with cursor-agent models.
- Effort: Cursor CLI has no effort levels.
- Session resume: The agent starts with the id of a chat opened by cursor-agent create-chat (--resume). If no id comes back, --continue is used.
- MCP: AgentVera’s MCP servers and browser are added as a local plugin and approved for the run.

**Antigravity**

- Sign-in: Antigravity uses its own Google login; if you’re signed out, it asks you to sign in in the browser on first launch.
- Model: Left empty, the model from your Antigravity settings is used. List models with agy models.
- Effort: Low, medium and high (--effort).
- Session resume: The agent opens its conversation by id (--conversation); without an id it continues the folder’s latest conversation (--continue).
- MCP: AgentVera’s MCP servers and browser are merged with your own mcp_config.json on macOS and Linux.

**opencode**

- Sign-in: Provider sign-in happens in opencode itself: /connect in the agent’s terminal or opencode auth login.
- Model: The model list comes from opencode models --verbose, grouped by provider and cached for 10 minutes.
- Effort: Each model’s own variants are shown as effort options.
- Session resume: Each agent starts in its own opencode session (--session ses_…). The session is created ahead of time by a short-lived opencode serve that only listens on 127.0.0.1; duplicating forks it the same way.
- MCP: The role, MCP servers and AgentVera browser travel as inline config and are merged with your own opencode.json.

**Hermes**

- Sign-in: Hermes uses the provider and credentials you picked with hermes model; if it isn’t set up yet, run hermes setup first.
- Model: Left empty, Hermes’ default model is used; AgentVera only passes an optional -m .
- Effort: Hermes has no effort levels.
- Session resume: The session id is read from the startup banner or, after the first reply, from hermes sessions list. If none is saved, the agent opens a new session.
- MCP: AgentVera’s MCP servers and browser are added to a managed Hermes config on every launch.

**Qwen Code**

- Sign-in: Qwen Code uses its own login, set with /auth in the agent’s terminal: an API key, a Qwen coding plan or a local model. The free Qwen OAuth tier ended on April 15, 2026.
- Model: Left empty, Qwen Code’s default model is used; a chosen model is passed with --model.
- Effort: Qwen Code has no effort levels.
- Session resume: Each agent starts with a session id from AgentVera (--session-id); a saved session opens with --resume.
- MCP: AgentVera’s MCP servers and browser are passed with --mcp-config.

**Kimi Code**

- Sign-in: Kimi uses its own login: kimi login, or /login in the agent’s terminal.
- Model: A model alias from Kimi’s config.toml; left empty, the default model is used.
- Effort: Low, medium, high and max.
- Session resume: Kimi creates the session on the first message; AgentVera finds its id with kimi session list and returns to it on later launches.
- MCP: AgentVera’s MCP servers and browser are passed with --mcp-config-file, next to your own mcp.json.

**Kiro**

- Sign-in: Kiro uses its own login (kiro-cli login); if you’re signed out, the agent’s terminal asks you to sign in.
- Model: Left empty, Kiro’s default model is used. List models with kiro-cli chat --list-models.
- Effort: Low, medium, high, extra high and max (--effort).
- Session resume: Kiro picks its own ids; when continuing, the agent opens with its session id or the folder’s latest session.
- MCP: AgentVera’s MCP servers and browser are merged with your own Kiro mcp.json on every launch.

**Amp**

- Sign-in: Amp uses its own ampcode.com login; if you’re signed out, the agent’s terminal asks you to sign in on first launch.
- Model: Amp uses modes instead of models; the mode decides the model, prompt and tools.
- Effort: Low, medium, high and ultra modes (--mode).
- Session resume: The agent attaches to a thread created with amp threads new and returns to it with threads continue on later launches.
- MCP: AgentVera’s MCP servers and browser are passed with --mcp-config.

**Droid**

- Sign-in: Droid uses its own Factory login; if you’re signed out, the agent’s terminal asks you to sign in at launch.
- Model: Left empty, Droid’s default model is used; the model is passed through a per-run settings file.
- Effort: Low, medium, high, extra high and max.
- Session resume: Droid writes its session file once the conversation starts; AgentVera finds it and returns to that session with --resume next time.
- MCP: AgentVera’s MCP servers and browser are merged on top of your Droid settings for that run.

**Goose**

- Sign-in: Goose uses the provider you picked with goose configure; it won’t open without one.
- Model: Left empty, the model chosen in goose configure is used; a chosen model is passed with --model.
- Effort: Low, medium, high and max.
- Session resume: Every agent opens a named session of its own (--name) and returns to it with --resume.
- MCP: AgentVera’s MCP servers and browser are added to Goose as extensions.

**Vibe**

- Sign-in: Vibe uses its own Mistral login or MISTRAL_API_KEY; a setup screen opens in the agent’s terminal on first launch.
- Model: A model alias from Vibe’s settings (VIBE_ACTIVE_MODEL); left empty, the default model is used.
- Effort: Vibe has no effort levels.
- Session resume: Vibe writes its session after the first message; AgentVera finds it and returns with --resume next time.
- MCP: AgentVera’s MCP servers and browser are merged by name with your own Vibe servers.

**Crush**

- Sign-in: Crush uses its own provider settings; the agent’s terminal asks on first launch.
- Model: Crush’s interactive mode has no model flag; model and provider are picked inside Crush (Ctrl+L) or in its settings.
- Effort: Crush has no effort levels.
- Session resume: The session is found with crush session list after the first reply; on the next launch the agent returns to it with --session.
- MCP: AgentVera’s MCP servers and browser are merged with your own crush.json on every launch.

**Aider**

- Sign-in: Aider reads provider API keys from environment variables or .aider.conf.yml.
- Model: Left empty, Aider’s default model is used; a chosen model is passed with --model.
- Effort: Low, medium and high (--reasoning-effort).
- Session resume: Each agent uses its own history file under ~/.aider/agentvera/ and continues with --restore-chat-history.
- MCP: AgentVera’s MCP servers aren’t passed to Aider.

**Cline**

- Sign-in: Cline uses its own provider setup (cline auth); if it isn’t set up, the agent’s terminal asks on first launch.
- Model: Left empty, Cline’s saved model is used; a chosen model is also written to Cline’s settings.
- Effort: Low, medium, high and extra high (--thinking).
- Session resume: Cline creates the session on the first message; AgentVera finds the id with cline history and returns with --id next time.
- MCP: AgentVera’s MCP servers and browser are merged with your own Cline MCP settings on every launch.

## Questions

**Which agents can use AgentVera’s MCP servers?** Claude Code, Codex, Gemini CLI, Copilot, Cursor, opencode, Hermes, Qwen Code, Kimi Code, Kiro, Amp, Droid, Goose, Vibe, Crush, Cline get AgentVera’s MCP servers and browser on every launch. Antigravity gets them on macOS and Linux only; Aider does not.

**Which agents have account chains?** Claude Code, Codex. When one account hits its limit, the agent carries on in the same conversation with the next account.

**Which agents let me pick an effort level?** Claude Code, Codex, Copilot, Antigravity, Kimi Code, Kiro, Amp, Droid, Goose, Aider, Cline. The others have no separate effort setting.

**How do I run a CLI that is not listed?** Any CLI that isn’t listed runs in a Terminal agent; the command only needs to be on your PATH.

---

**About AgentVera.** AgentVera is a desktop app for macOS, Windows and Linux, built by Solviera Teknoloji. It runs 18 AI coding agents (Claude Code, Codex, Gemini CLI, Copilot, Cursor, Antigravity, opencode, Hermes, Qwen Code, Kimi Code, Kiro, Amp, Droid, Goose, Vibe, Crush, Aider, Cline) side by side, per project, in real terminals with the user's own CLI installs and logins. The free plan has unlimited agents. Download: https://agentvera.dev/en/download/ · Pricing: https://agentvera.dev/en/pricing/