AI agent, MCP, and CLI

MovieClaw builds a general-purpose agent into the media server. Say what you want, and it searches, subscribes, upgrades, tidies up your library, and generates subtitles, showing every tool call as it goes. The same abilities are open to the outside, so any agent can do everything the web app can through MCP or the mclaw CLI.

The built-in agent

1. Add a model provider under Settings → Model providers (模型接入). OpenAI, Alibaba Cloud Model Studio, DeepSeek, Zhipu GLM, and Kimi are preset, and any OpenAI-compatible endpoint works too (self-hosted vLLM, SiliconFlow, OpenRouter, and so on). You can add several at once.
2. Under Settings → AI (AI 设定), choose the default model for the agent and for subtitles.
3. Go back to the home screen and just ask, like "subscribe to The Three-Body Problem in 4K" or "clean up the messy file names in library 1".

The guardrails live in code: credentials are never handed to shell tools, destructive actions need your confirmation, and deleted files go to a recycle bin first.

From your chat apps

Under Settings → Messaging (消息推送), connect WeChat, Telegram, or Discord to chat with the agent and get notifications there. Feishu currently gets notifications only. Conversations coming in from chat apps get a restricted set of tools.

Outside agents: MCP

1. As an admin, create an endpoint under Settings → MCP server (MCP 服务) and pick which abilities it exposes. Each endpoint has its own token that you can revoke anytime.
2. Give your agent the endpoint URL and token. The MCP server page shows the full URL ready to copy. With Claude Code it looks like this:

claude mcp add --transport http movieclaw http://192.168.1.10:3000/mcp/<endpoint> \
  --header "Authorization: Bearer <token>"

Cursor, Cline, and custom agents take the same URL and Authorization header in their config. Endpoints use MCP's Streamable HTTP transport with static tokens, so web connectors that require OAuth can't connect yet.

The mclaw CLI

mclaw is a single static binary with no Python or Node needed. It's already in the server image (docker exec -it movieclaw mclaw status). On other machines, install it with one command:

# Linux, macOS
curl -fsSL https://raw.githubusercontent.com/movieclaw/movieclaw/main/scripts/install-cli.sh | sh

# Windows (PowerShell)
irm https://raw.githubusercontent.com/movieclaw/movieclaw/main/scripts/install-cli.ps1 | iex

Then pair it and go:

mclaw login --server http://192.168.1.10:3000   # pair; on your LAN you can omit the address
mclaw status                                    # server and auth status
mclaw search titles "Dune"                      # search titles
mclaw subscriptions list                        # list subscriptions
mclaw subscriptions --help                      # every command has --help

mclaw login shows a pairing code. Approve it in the web app under Settings → Devices (设备). The command tree is generated from the server's API, so every new backend feature shows up in the CLI automatically. Output is JSON in pipes and agents, and destructive commands require --yes. Any agent that can run a shell can drive MovieClaw with it.