Kimi Code CLI is Moonshot AI's terminal coding agent. It runs locally in a project, reads and edits code, executes shell commands, searches files, fetches web pages, and iterates based on tool feedback. Moonshot distributes it through official install scripts, Homebrew, and npm, with the kimi executable as the primary terminal entry point.
Key capabilities
Terminal coding workflow - Kimi Code can inspect a repository, modify files, run tests or build scripts, and explain what it changed. The official docs show a supervised model where read-only operations run automatically by default, while file modifications and shell commands ask for confirmation.
Automation surfaces - The CLI supports kimi -p for one-shot prompts and kimi -C for resuming the previous session. Moonshot also publishes the Kimi Agent SDK, which exposes the Kimi CLI runtime to Go, Node.js, and Python applications while preserving CLI configuration, tools, skills, MCP servers, approvals, and streaming responses.
MCP, plugins, and subagents - Official sources describe conversational MCP configuration, a plugin/skill marketplace, lifecycle hooks, and built-in coder, explore, and plan subagents. Kimi also exposes kimi acp for Agent Client Protocol editor integration with Zed, JetBrains, and other ACP clients.
Autonomy level
Level 3: a supervised coding agent. It can plan and execute multi-step work, but the current public docs still emphasize confirmations for writes and shell commands.
Strengths
- Open-source terminal agent with current official docs and active Moonshot AI ownership
- Scriptable prompt mode plus SDK path for automation and orchestration
- MCP, skills/plugins, hooks, subagents, and ACP give it a broad integration surface
- First-party Kimi model path, with configuration hooks for compatible providers
Limitations
- New and fast-moving, so command flags and config paths need frequent re-verification
- Windows usage depends on Git Bash according to the official docs
- DevThrottle still needs empirical terminal-mode, session-store, and transcript-export testing