PlugWiki
Z.AI Vision MCP logo

Z.AI Vision MCP Server: Setup for Claude Code, Cursor and Other Apps

OfficialLocal
Getnpm@z_ai/mcp-server
Tools
8
8 read · 0 change
Downloads
20.8K
a week
Works in
12
apps
Risk
Low
read only
Auth
API key

Lets Claude, Cursor and other assistants that cannot see images read screenshots, diagrams, charts and short videos through Z.AI.

Z.AI Vision MCP is made by Z.AI. It adds eight image and video tools to your assistant, and Z.AI's vision model does the looking. You can turn a UI screenshot into code, read text from a screenshot, explain an error or diagram, check a chart, compare two screens, or describe a short video.

It runs on your computer and needs an API key. Z.AI describes it as made for GLM Coding Plan users and points to the Coding Plan page for the key; a Zhipu BigModel key also works. Set the mode to match where your key comes from. Without it, the server assumes the BigModel platform and a Z.AI key will not work.

Every image or video you point it at goes to Z.AI's servers for analysis. Images over 5 MB and videos over 8 MB are rejected.

Good for

  • Developers who want code from a UI screenshot or a design
  • Anyone who debugs from error screenshots and terminal captures
  • Users of text-only coding models who need image and video understanding

Works in

12 should work

Tap an app to see how to set it up and any known problems.

Install Z.AI Vision MCP

Claude Code

  1. 1

    Copy this prompt.

    Add the Z.AI Vision MCP server to Claude Code. Run this command:

claude mcp add z-ai-vision -e Z_AI_API_KEY=<Z_AI_API_KEY> -e Z_AI_MODE=<Z_AI_MODE> -- npx -y @z_ai/mcp-server

Ask me for Z_AI_API_KEY, Z_AI_MODE and only put the values in that config, nowhere else.

Then check that it loaded and tell me if I need to sign in or restart.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

Built from the package data we hold. Not yet tested by us.

Tools

  • Turns a UI screenshot into code, a prompt or a spec

    Takes a screenshot of an interface and turns it into front-end code, a prompt for recreating it, a design spec, or a plain description. Sends the image to Z.AI's vision model.

    Try: “Turn the screenshot at ~/Desktop/login.png into React code.”

    Files · ui_to_artifact

  • Reads text out of a screenshot

    Reads text from a screenshot, such as code, terminal output or documentation. You can give a programming language hint to keep formatting right.

    Try: “Extract the code from ~/Desktop/snippet.png as Python.”

    Files · extract_text_from_screenshot

  • Explains an error screenshot

    Looks at a screenshot of an error message or stack trace, names the likely cause and suggests fixes and ways to avoid it.

    Try: “Why does the error in ~/Desktop/crash.png happen?”

    Files · diagnose_error_screenshot

  • Explains an architecture or flow diagram

    Explains technical diagrams such as architecture drawings, flowcharts, UML, ER and sequence diagrams.

    Try: “Walk me through the architecture diagram at ~/Desktop/arch.png.”

    Files · understand_technical_diagram

  • Finds insights in a chart or dashboard

    Reads charts, graphs and dashboards from an image and reports trends, anomalies and what they may mean for the business.

    Try: “What trends does the chart in ~/Desktop/sales.png show?”

    Files · analyze_data_visualization

  • Compares two UI screenshots

    Compares an expected screenshot with the real one and lists the visual differences, ranked by how serious they are.

    Try: “Compare design.png and build.png and list what differs.”

    Files · ui_diff_check

  • Answers any question about an image

    A fallback for images the other tools do not fit. You give an image path or URL and a prompt describing what you want to know.

    Try: “Describe what is in https://example.com/photo.jpg.”

    Files · analyze_image

  • Describes what happens in a short video

    Analyzes a short video from a local file or web address, up to 8 MB in MP4, MOV or M4V, and answers your prompt about it.

    Try: “Summarize what happens in ~/Movies/demo.mp4.”

    Files · analyze_video

Read from the source code

Permissions

low risk · 3 things to watch

Low risk

The tools only read images and videos, but every file you give them is uploaded to Z.AI for analysis. Do not point them at screenshots that contain secrets or private data.

What it can do

Watch out for

  • Your images and videos leave your computer

    Every file or URL you give a tool is sent to Z.AI, or to the BigModel platform, for analysis. Do not use it on screenshots with passwords, customer data or private code. npmjs.com

  • It reads any local file path it is given

    The tools read image and video files from any path the assistant names, then upload them. Check which file the assistant picks before approving a call. npmjs.com

  • Wrong mode sends your key to the other platform

    Z_AI_MODE decides whether requests go to api.z.ai or open.bigmodel.cn, and the default is the BigModel platform. Set ZAI for a z.ai key. npmjs.com

Security advisories

No published security advisories as of Nov 1, 2026.

Technical details
Turning on read-onlyAll eight tools only read an image or video and return text. There is no write or delete tool, so no switch is needed.
Talks toapi.z.ai or open.bigmodel.cn, depending on Z_AI_MODE (your images and videos are sent there)
Sign-in and keysYour Z.AI or BigModel API key is set in the server's environment variable Z_AI_API_KEY in your app config.

Questions

Is the Z.AI Vision MCP server official?

Yes. It is published by Z.AI under the @z_ai package scope, and Z.AI documents it in its own developer docs.

Does it need an API key?

Yes. Z.AI documents the server as made for GLM Coding Plan users and says to take the key from your Coding Plan page, so it is tied to that paid plan. A Zhipu key from bigmodel.cn also works. Set the mode to ZAI for the first and ZHIPU for the second. Z.AI's page asks for Node.js 22 or newer.

What does Z_AI_MODE do?

It picks which platform your requests go to. ZAI uses api.z.ai and ZHIPU uses open.bigmodel.cn. If you leave it out, the server uses ZHIPU, so a z.ai key fails with an authentication error.

Does it send my screenshots to the cloud?

Yes. Images and videos are sent to Z.AI for analysis. Local files are encoded and uploaded, and web addresses are passed on.

What are the file size limits?

Images up to 5 MB and videos up to 8 MB in MP4, MOV or M4V format. Larger files are rejected.

Which AI apps does it work with?

The README shows Claude Code and points to the Z.AI docs for other apps. It runs locally over stdio, so any app that starts local servers can use it. Follow the install box.

Can it change or delete anything?

No. All eight tools only read an image or video and return text.

Information

updated Nov 2, 2026

Publisher
Official
Package
@z_ai/mcp-server
Registry
npm
Transport
stdio
Auth
API key
Version
0.1.6
License
Apache-2.0

Websitenpm

Sources

Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.

Show all 2
  1. www.npmjs.com · npmjs.com
  2. docs.z.ai · docs.z.ai

Facts from the official MCP registry, GitHub, npm, PyPI and the Docker catalog. Updated Nov 2, 2026.