Vizro MCP

What mckinsey/vizro-mcp does

mckinsey/vizro-mcp connects an LLM host to Vizro through the Model Context Protocol. Its purpose is to help generate dashboards and charts, using tools and templates designed around validated and maintainable Vizro outputs.

The server is intended for data visualization workflows where an LLM can assist with creating or refining a dashboard. It is useful when the target application is a Vizro dashboard rather than a general-purpose charting workflow.

The server operates in a read-only manner. It can read data needed for visualization tasks, but it does not write, delete, or modify data on the local machine.

How it works

An MCP-compatible LLM application starts the server and makes its visualization capabilities available to the model. The model can then work with Vizro dashboard and chart creation tasks through the MCP connection.

The server can be launched as a Python tool with uvx or as a Docker container. In either case, it is configured as an MCP server in the host application's configuration file. The host application remains responsible for the interaction with the model and for actions taken through that application.

When Docker is used with local data, the required files or directories must be mounted into the container. The mount configuration uses absolute paths and can expose either a directory or an individual data file to the container.

Setup and configuration

Two setup routes are documented:

• With uvx, install uv, confirm its location with which uv, and configure the returned uvx path as the MCP command with vizro-mcp as its argument.
• With Docker, install Docker and configure the host to run docker run -i --rm mcp/vizro.

Claude Desktop, Cursor, and VS Code have dedicated setup guidance. Cursor and VS Code also provide quick-install options for both the uvx and Docker approaches. Docker users working with local data need to add bind mounts for the relevant directories or files.

The LLM application must already be installed and able to act as an MCP host. Model choice affects the quality and consistency of generated results; the project recommends a high-performing model and specifically mentions claude-4-sonnet.

Tools and capabilities

• Create Vizro dashboards with LLM assistance.
• Create data charts through MCP.
• Use templates for dashboard and chart generation.
• Produce validated, maintainable visualization outputs.
• Read local data without changing it.

Limitations and notes

The server is read-only with respect to data on the machine. It does not write, delete, or modify local data. Access to local files in a Docker deployment depends on explicitly mounting the relevant paths.

The LLM host and its users remain responsible for activity performed through the host application. Results generated by third-party generative AI tools may contain inaccuracies, bias, misleading content, or other unsuitable material. Users are also responsible for obtaining any rights needed to use third-party generative AI services and for following their applicable terms.