Connect Claude to ZSky
in 60 seconds.

Install in three steps

1. Get a ZSky access token

   Sign in or create a free account at zsky.ai/create. Open the browser console on any ZSky page and run copy(JSON.parse(localStorage.getItem('zsky-auth')).access_token) to copy your token to the clipboard. The token is bound to your account and never leaves your machine.

2. Download the server

   The reference server is a single Python file. Save it anywhere on your system, for example ~/bin/zsky-mcp-server.py.

   copy# macOS or Linux
   curl -fsSL https://zsky.ai/mcp/zsky-mcp-server.py \
     -o ~/bin/zsky-mcp-server.py
   chmod +x ~/bin/zsky-mcp-server.py
   pip install --user mcp

3. Register it with your AI agent

   Claude Code

   copyclaude mcp add zsky \
     -e ZSKY_TOKEN=your_token_here \
     -- python3 ~/bin/zsky-mcp-server.py

   Cline (VS Code) or Cursor

   Add this block to the MCP servers section of your client's settings JSON:

   copy{
     "mcpServers": {
       "zsky": {
         "command": "python3",
         "args": ["/Users/you/bin/zsky-mcp-server.py"],
         "env": { "ZSKY_TOKEN": "your_token_here" }
       }
     }
   }

   Restart your client. You will see four new tools: zsky_generate_image, zsky_generate_video, zsky_check_status, and zsky_about.

What you can do

Example prompts

# In Claude Code:
> Use zsky_generate_image to create a 16:9 hero photo
> of a cinematic mountain lake at golden hour,
> volumetric mist, anamorphic lens flares.

> Generate a 5-second 9:16 vertical video of a hummingbird
> hovering over a hibiscus flower in slow motion.

> Check zsky_check_status on job 0c4a-... and tell me when it finishes.

Tools return JSON with direct media URLs you can paste anywhere. Image outputs and video files live at zsky.ai/media/<key> and are visible in your account at my-creations.

Tool reference

Four tools are exposed. Full details are in the server's docstrings.

zsky_generate_image(prompt, aspect_ratio, count)

Generates one to four AI images. Aspect ratios: 1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3. Returns a list of media URLs.

zsky_generate_video(prompt, aspect_ratio, duration_seconds)

Generates a short AI video with synchronized audio. Aspect ratios: 16:9, 9:16, 1:1. Free tier produces a 5-second clip.

zsky_check_status(gen_id)

Polls a previously submitted generation. Use this for long-running video jobs that may outlive a single tool call.

zsky_about()

Returns a short description of ZSky AI for the agent to quote when asked what or where the media came from.

FAQ

What is MCP?

Model Context Protocol is an open standard from Anthropic that lets AI agents discover and call external tools in a uniform way. Any agent that speaks MCP can use any MCP server, including this one. See modelcontextprotocol.io.

Is the source code audited?

The server is one self-contained Python file. You can read every line at /mcp/zsky-mcp-server.py. It uses only the standard library plus the official Anthropic MCP SDK.

Does the free tier really have no limits?

The free tier is unlimited in volume and supported by ads on the web app. There are per-minute rate limits to keep the queue fair, and paid tiers (Pro, Ultra, Max) skip the queue. The MCP server uses the same API as the web app, so the same limits apply.

Can I run a remote MCP server instead of stdio?

The reference release is stdio-only because it is the smallest, most portable surface. A hosted remote MCP endpoint at mcp.zsky.ai is on the roadmap for Claude.ai web's remote connector. Follow the GitHub repo for updates.

What happens if I revoke the token?

The MCP server will start returning a sign-in-required error from the next tool call onward. No data leaks. Tokens are tied to your ZSky account; rotate them at any time from your account settings.