← Back to Discover

Kimi Code

Moonshot AI's terminal coding agent, with video input, MCP setup by conversation, subagents, lifecycle hooks and editor integration over the Agent Client Protocol.

Coding agentsCommand-line tools

What it does

Kimi Code CLI is a coding agent for the terminal. It can modify code, execute shell commands, look through files and pull in web pages, then decide its next step from the results. It works with Moonshot AI's Kimi models out of the box and can be set up with other compatible providers. After installing, you run kimi in a project and log in with Kimi Code OAuth or a Moonshot AI Open Platform API key. The README highlights a single-binary install through a script for macOS, Linux and Windows PowerShell, fast startup, and video input so you can give the agent a screen recording or demo clip. MCP servers are added and authenticated conversationally with /mcp-config. Built-in coder, explore and plan subagents work in isolated contexts, and lifecycle hooks can gate risky tool calls or connect to your own automation. Through the Agent Client Protocol, the kimi acp command lets editors such as Zed and JetBrains drive a session. Skills, MCP servers and data sources can be installed from a marketplace or any GitHub repo. The project is MIT licensed.

Capabilities

  • Terminal interface with fast startup and single-binary install
  • Video input from screen recordings or demo clips
  • MCP servers configured conversationally with /mcp-config
  • Coder, explore and plan subagents
  • Lifecycle hooks to gate risky tool calls
  • Editor integration with Zed and JetBrains over ACP

When to use it

  • You want a terminal coding agent that works with Kimi models or other compatible providers
  • You want to show the agent a screen recording or demo clip
  • You want to drive an agent session from Zed or JetBrains
  • You want hooks that gate risky tool calls