A Model Context Protocol (MCP) server and app for creating Datawrapper charts using AI assistants. Built on the datawrapper Python library.

Status: early and experimental. This project — including its Claude Code/Desktop plugin and marketplace listing — is under active development, and behavior can change between releases. In particular, the plugin/marketplace install path has known gaps (see issue #78) — don't treat it yet as a finished, zero-maintenance install path for a wider rollout.

You can provide a data file and simply ask for the chart you want. The draft will soon appear in the panel.

Here's a more complete example showing how to create, publish, update, and display a chart by chatting with the assistant:

"Create a datawrapper line chart showing temperature trends with this data:
2020, 15.5
2021, 16.0
2022, 16.5
2023, 17.0"
# The assistant creates the chart and returns the chart ID, e.g., "abc123"

"Publish it."
# The assistant publishes it and returns the public URL

"Update chart with new data for 2024: 17.2°C"
# The assistant updates the chart with the new data point

"Make the line color dodger blue."
# The assistant updates the chart configuration to set the line color

"Show me the editor URL."
# The assistant returns the Datawrapper editor URL where you can view/edit the chart

"Show me the PNG."
# The assistant embeds the PNG image of the chart in its contained response.

"Suggest five ways to improve the chart."
# See what happens!

bar, line, area, arrow, column, multiple column, scatter, stacked bar

Use list_chart_types to see descriptions, then get_chart_schema to explore configuration options for any type.

• A Datawrapper account (sign up at https://datawrapper.de/signup/)
• An MCP client such as Claude or OpenAI Codex
• Python 3.10 or higher

1. Go to https://app.datawrapper.de/account/api-tokens
2. Create a new API token
3. Add it to your MCP configuration as shown in the installation guide

{
  "mcpServers": {
    "datawrapper": {
      "command": "uvx",
      "args": ["datawrapper-mcp"],
      "env": {
        "DATAWRAPPER_ACCESS_TOKEN": "your-token-here"
      }
    }
  }
}

For other clients (Claude Desktop, Claude.ai, Cursor, VS Code Copilot, ChatGPT, OpenAI Codex, OpenClaw) and Kubernetes deployment, see the installation guide.

Installing via a plugin marketplace/directory (Claude Desktop's plugin browser, ClawHub, etc.) is still experimental. These interactive install flows currently have no working way to collect required environment variables like DATAWRAPPER_ACCESS_TOKEN — see issue #78 for details. The manually-edited config shown above (and throughout the installation guide) is the reliable path today.

When connecting to a hosted instance of the server over HTTP, you can authenticate with your own Datawrapper API token by sending it in the Authorization header:

Authorization: Bearer <your-datawrapper-api-token>

This ensures charts are created under your account instead of the server operator's. The token is read from the header automatically — no need to include it in every tool call.

You can also pass access_token directly as a tool argument, which takes precedence over the header. When neither is provided, the server falls back to its DATAWRAPPER_ACCESS_TOKEN environment variable.

Set DATAWRAPPER_MCP_INSTRUCTIONS to have the server hand your own free-text guidance to connecting MCP clients (many, including Claude, fold this into the model's context). Use it for house style rules — required fields, naming conventions, and the like — without forking the server:

"env": {
  "DATAWRAPPER_ACCESS_TOKEN": "your-token-here",
  "DATAWRAPPER_MCP_INSTRUCTIONS": "Every chart needs alt text and a CMS slug in its notes field."
}