# CLI 0.5: 18 coding agents

> 2026-10-19 · CLI 0.5.0 · New, Fixed, CLI, Agents · https://comity.so/changelog/2026-10-19-cli-0-5

`comity init` now sets up 18 coding agents, so everyone on the team can keep the agent they like. Maya on Cursor, Jordan on Codex, Priya on Gemini CLI and Diego on Claude Code all work in the same repo, and all of them hear about each other's changes.

## New

### 18 agents

- **Hooks and MCP**: Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Kiro, Factory Droid, Cline, OpenCode, Amp, goose, Junie, Augment and omp. Comity runs in the agent's own hooks, so it hears about edits as they happen, and the agent gets Comity's tools over MCP.
- **MCP and instructions**: Roo Code, Continue and Zed, which have no hooks. Their agents call Comity's tools when the instructions tell them to.
- Any other MCP client can connect to the MCP server by hand.

```sh
comity init --tools cursor,gemini   # pick agents by id
comity init --tools all --yes       # every agent, no questions
```

Run plain `comity init` to choose from a list with the agents on your machine pre-selected. Every agent has its own docs page with exactly what init writes, for example [Cursor](/docs/agents/cursor) or [Gemini CLI](/docs/agents/gemini).

### `comity mcp`

A small local MCP server that forwards your agent's requests to Comity, signed in with your `comity login`. Agents that can't send a login header use it, so no token is ever written into a config file. See [Stdio bridge](/docs/agents/mcp-clients#stdio-bridge-comity-mcp).

## Improved

- **The same committed files on every OS.** Hook and MCP entries start `comity` through a one-line Node launcher, so a teammate on Windows and one on macOS commit identical files. Node needs to be on your `PATH`, which the CLI needs anyway.
- The dashboard shows each agent's logo.

## Fixed

- Hooks on Windows no longer fail when the input an agent sends starts with a byte order mark.