About datatomcp

datatomcp is a hosted service that turns a CSV or Google Sheet into a remote MCP server and REST API, so AI agents like Claude, ChatGPT and Cursor can filter, group and total the data with exact, versioned results.

Who builds it

datatomcp is built and run by Killian Helf, founder, from The Hague, the Netherlands. It is a small, independent product: the person who writes the code also answers the email. Write to [email protected] and you get a reply from the founder.

Why it exists

Agents are good at deciding what to ask and bad at reading a 50,000-row spreadsheet pasted into a chat. Pasted data costs context, slows answers down, and turns totals into estimates. datatomcp keeps the data in a database (DuckDB) and gives the agent a few validated query tools instead, so the model asks and the database counts.

How it's built

• Exact numbers. Counts, sums and averages are computed by the database, and every result names the dataset version it came from.
• Read-only by design. Agents can query your data but never change it, and they can't run raw SQL.
• Nothing silent. Every import comes with a report of what was found: bad lines by number, numbers stored as text, mixed date formats. Fixes happen only when you confirm them.
• Your data stays yours. Stored in the EU, never sent to AI providers, never used for training, deleted when you delete it.
• Open standards. One MCP URL with OAuth works in any client that supports remote MCP servers, and the same engine serves a plain REST API.

Get in touch

Questions, bugs, feature requests or a file that doesn't import the way you expect: [email protected].