AWS launches open-source MCP server to make its open data catalog searchable by AI

Making open data findable

The Registry of Open Data on AWS (RODA) hosts over 1,100 datasets from more than 400 organizations, including NASA, NOAA, NIH and the Allen Institute. Through its Open Data Sponsorship Program, AWS covers the storage costs, making more than 100 petabytes of satellite imagery, climate records, genomics data and geospatial research freely available.

According to Tonix Alejo, AI for Good Lead at AWS, the problem is discovery. A climate researcher studying drought in East Africa might spend days finding the right dataset among 1,100 options, working out the file structure and judging whether it fits the question. The new server is meant to let a researcher simply ask for, in the example given, "datasets related to drought indices in sub-Saharan Africa" and receive options, previews and metadata in seconds.

How the MCP server works

MCP is an open standard that lets AI assistants connect to external tools and data sources. With the server, any MCP-compatible assistant, including Kiro and Claude Code, can search the full catalog, inspect metadata, list bucket contents and sample files on the researcher's behalf.

The server is open source under the Apache 2.0 license and is built around three capabilities:

• Discovery: finding datasets by research question.
• Exploration: inspecting metadata and finding related datasets.
• Evaluation: previewing file structures and sampling actual data.

At the Summit panel Open data for all, Dan Pilone, CEO of Element 84, said researchers and emergency responders need the right data delivered quickly, reliably and in a form that supports action. David Feng of the Allen Institute added that high-quality open datasets generate value well beyond the original research project, helping scientists test methods and build on earlier discoveries.

The Now Go Build CTO Fellowship

AWS paired the launch with its Now Go Build CTO Fellowship, led by Amazon.com VP and CTO Werner Vogels. The program supports technology leaders working on social impact in education, health, food security, climate resilience and disaster response, often with limited budgets and high stakes. Vogels said: "We feel, as a large organization, responsible for helping nonprofits to actually be successful with technology." He encouraged engineers and business leaders to give time and expertise as well as funding.

Nineteen Fellows took part in this year's Summit as speakers, invited through a collaboration between AWS and the International Telecommunication Union (ITU). Fred Werner of ITU, after a mentorship session with them, said: "They're not talking about AI in theory. They're building practical solutions to real problems, often local solutions for local communities."

Season 2 of the Fellowship's video series premiered at the Summit and follows Fellows from the Health and Education cohorts:

• Ezekiel Brooks of U.S. Hunger uses AI and data integration to uncover root causes of food insecurity and connect families to healthcare and social services.
• Jay Patel of Jacaranda Health delivers maternal health guidance by SMS to three million mothers across Africa.
• Oluwafunmilayo Ajakaiye of The Sapphires Development Initiative uses AI-powered personalized learning to bring STEM education to underserved girls across Africa.
• Rosius Ndimofor of EduCloud Academy in Cameroon builds cloud platforms in local languages.

Other speakers described local data gaps. Petya Kangalova of the Humanitarian OpenStreetMap Team said local contributors fill holes in mapping data for disaster response, since satellite imagery and global models are often incomplete for a given place. Vijay Solanky of the Watershed Organisation Trust said smallholder farmers in India decide using models that lack accurate local soil, water and crop information, and his team builds that missing data layer with communities.

What it means for researchers and builders

AWS describes the two efforts as addressing different sides of one problem: the Fellowship invests in people who understand local constraints, while the MCP server addresses infrastructure by making scientific data discoverable and usable. According to AWS, researchers can find the right dataset in minutes rather than days, and builders on the ground get data, tools and support to turn local knowledge into working systems.

The MCP server for RODA is open source and can be added to any MCP-compatible AI assistant via its GitHub repository. Organizations interested in the Fellowship can learn about the nomination process.