There are millions of datasets scattered across government portals and tools, accessible mainly via APIs or CSVs. Different formats, different schemas, no easy way to merge them.
I've been building autario to fix this. It's a data layer that makes public datasets queryable — for humans, apps, and AI agents.
What it does
- 2,500+ datasets and counting from World Bank, IMF, Eurostat, FRED, SIPRI, IEA, NASA, NOAA
- REST API; filter, sort, paginate any dataset by URL
- MCP support; Claude, ChatGPT, and any AI agent can search, query, and publish charts without hallucinations
- Interactive charts; 100+ visualizations and counting
The MCP part
This is what I'm most excited about. Instead of LLMs hallucinating numbers, they can query autario in real-time:
{
"mcpServers": {
"autario": {
"command": "npx",
"args": ["autario-mcp"]
}
}
}
12 tools: search_datasets, query_dataset, publish_chart, etc. The LLM searches for data, queries it, and if you wish, can publish a chart for future use. All with verified numbers. You can manually edit the charts later on.
Example: Military Spending
I asked Claude to compare military spending across countries. It searched autario, found the SIPRI dataset, queried the data, and published this:
View interactive chart on autario
The automatically generated insight text uses real numbers from the data, not from the model's training data, but can be manually edited too.
What I'd love feedback on
I'm trying to figure out what to build next:
- What datasets do you wish were easier to access? Public? Private? Which ones?
- Would you use this in your workflow? (dashboards, ML training, research, journalism?)
- Would private/proprietary datasets be valuable? Thinking about letting users upload and sell their own data
- What's the biggest pain point when working with data?
The platform is free at autario.com. API docs at autario.com/documentation. npm package: npx autario-mcp.
Would love to hear what would make this actually useful for your workflow.
Thanks a lot!
Niko
Top comments (0)