OpenGenerativeUI MCP
The OpenGenerativeUI MCP server wraps HTML fragments into a styled document and gives an agent skills for SVG diagrams
- 4,227 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Returns HTML with scripts that must not be shown outside an isolated frame
- Open CORS by default
Install
- Works in
Manual install
{
"openGenerativeUI": {
"url": "http://localhost:3100/mcp"
}
}An entry for `.mcp.json`. Start the server first with `pnpm dev` in `apps/mcp`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Returns HTML with scripts that must not be shown outside an isolated frame
- Open CORS by default
This is third-party code. Review the repository files before installing.
What it does
OpenGenerativeUI is an open generative UI framework from CopilotKit. The `apps/mcp` directory holds a separate MCP server that can be deployed independently of the rest of the project. The `assemble_document` tool wraps an agent's HTML fragment in a full document with a ready design system and bridge JavaScript so the result renders straight in a frame. Through the `skills://list` and `skills://{name}` resources the agent reads instructions for visualization and SVG diagrams. Prompt templates `create_widget`, `create_svg_diagram` and `create_visualization` are included.
Who it is for. Developers having an agent draw widgets and diagrams who want one styled result instead of bare HTML.
Good fit when
- An agent has to draw an interactive chart and the answer must show in an iframe.
- An architecture SVG in one style is needed for documentation.
- Several agents produce visualizations and all must look alike.
- A widget prototype is built right in a model chat.
- An HTTP MCP client should receive a styled document on request.
Not a fit when
- You need a ready production package: the server is built from source and the README names no ready package.
- The result is shown without isolation in a frame.
- A plain Markdown answer with no visuals is enough.
Example request
Build an interactive binary search visualization and wrap it with assemble_document so it can be shown in a frame.MCP
- Transport
- stdio, http
- Authentication
- not required
Environment variables
MCP_PORT- HTTP server port, default 3100
ALLOWED_ORIGINS- Comma-separated CORS origins, any by default
Limitations
The server runs from a clone: `pnpm install`, then `pnpm dev` on port 3100, or a build and `dist/stdio.js` for stdio clients. The README names no published package. CORS allows any origin by default (`ALLOWED_ORIGINS=*`), so narrow it outside development. The tool returns HTML with scripts, so show it only in an isolated frame. The last commit on the main branch is from October 2026.
How to disable. Remove the `open-generative-ui` entry from `claude_desktop_config.json` or `.mcp.json` and stop the server process.
FAQ
How do I start the server?
In `apps/mcp` run `pnpm install`, then `pnpm dev`. The server opens at `http://localhost:3100/mcp` and the health check answers at `/health`.
How do I connect Claude Desktop?
Build with `pnpm build` and add an entry to `claude_desktop_config.json` with the command `node`, the argument `dist/stdio.js` and a `cwd`.
What does assemble_document accept?
A title, a one-sentence description and a self-contained HTML fragment with its own styles and scripts. A full document for an iframe comes back.
What can be configured?
The port `MCP_PORT`, CORS origins `ALLOWED_ORIGINS`, skills directory `SKILLS_DIR` and log level `LOG_LEVEL`.
Where do the skills come from?
The `skills://list` resource returns the list and `skills://{name}` returns the instruction text.
Discussion
No questions or reviews yet.
Related
MCP serversA MasterGo MCP server that hands agents the DSL of a design file from a link so they rebuild the layout as HTML or a component.
A plugin with skills for building TypeScript MCP servers and MCP apps: the agent scaffolds, edits, debugs and verifies them
GolemUI MCP server and skill validate declarative forms and generate them from JSON Schema or OpenAPI.
Fluid is a generative-art MCP server that builds backgrounds and wallpapers from 24 math fields and returns embed code for a site.
An MCP server for generating images and video from the agent: several models, a large prompt library and parallel tasks
OpenPencil's MCP server: agents inspect designs, edit layers and export assets, including .fig files, without Figma.
README in short
The server README describes a standalone package with no dependency on the rest of the repository, HTTP and stdio start-up and configuration through environment variables.