datatomcp
Hosted data layer that lets AI agents query a team's sheets, databases and apps with exact, audited answers
100X Meter Early
Think datatomcp could go 100X?
Give it a push. One tap per reader each day; the meter tracks buzz here, not quality.
Tell your network you backed datatomcp early
Download share cardWhat datatomcp does
datatomcp is a hosted service that sits between a team’s data and the AI agents the team uses. You connect Google Sheets, CSV or Excel files, a PostgreSQL or MySQL database, or apps such as Stripe, HubSpot and Airtable. Each source becomes a dataset with versions. Agents such as Claude, ChatGPT, Cursor or Claude Code then ask questions through one remote MCP server, or through a REST API for tools without MCP support. The site calls it an “AI-native data layer for teams without a data team”.
The problem it targets is a common one. Someone pastes a large spreadsheet into a chat, the context window fills up, the token bill grows and the totals the model gives back are close but not exact. datatomcp’s answer is to let a database do the counting. The agent asks for a total or a filtered set of rows and gets back only the result, along with the version of the dataset it came from. The company says that on its own 50,000-row test file, a full session sent about 475 times fewer tokens than pasting the whole file. That is its own measurement, and we haven’t repeated it.
How it works
The site describes five steps. Connect: add a source, and an import report lists rows it couldn’t read, numbers stored as text and mixed date formats, without changing anything until a person confirms. Link: it looks for ids that datasets share, such as customer_id in orders and in customers, so an agent can total orders by a customer’s region in one query. A link only confirms itself when every value matches. Define: it drafts metric definitions from your columns, for example paid revenue as the sum of amount where status is paid, and a person approves or edits each one. Agents only use approved definitions, and a glossary covers team words like “active customer”. Ask: agents query through read-only tools. Audit: every agent question is logged with who asked, which agent, the query, how long it took and any error, and repeated questions are compared to show whether two agents got the same answer.
Agents can’t run raw SQL. They get a fixed set of validated tools (list and describe datasets, query rows, aggregate, draw a chart, make a dashboard, flag an issue) and results are capped at 1,000 rows per call. The charts come from the company’s own open-source library, published on npm as @datatomcp/charts. An optional in-app assistant runs on your own AI key, and optional AI clean-up suggestions send a column summary and up to 50 sample rows to Mistral in the EU, only when an editor asks.
Data and privacy
The site says data is stored in the EU, never used for training, and deleted at once when you delete a dataset (and from backups within 14 days). Database and app credentials are encrypted and not shown to agents. Private Google Sheets are picked through Google’s own file picker, so the service gets access to those files only.
Who makes it
The site says datatomcp is built and run by its founder, Killian Helf, in The Hague in the Netherlands. In a note on the home page he says he helps finance staff at a large company use AI agents and kept seeing them hit the paste-the-spreadsheet wall. The product was announced on 4 November 2026 and posted to Hacker News as a Show HN on 12 November 2026.
We haven’t tested datatomcp ourselves. It launched recently, so its features, limits and prices may change. Everything on this page comes from the official site and the public pages linked in the sources, checked on 15 November 2026.
What people use datatomcp for
- Let Claude or ChatGPT answer revenue questions from a Google Sheet without pasting the sheet into the chat
- Give several agents on one team the same approved definition of a metric such as paid revenue
- Join orders, customers and regions kept in different sources and total them in one query
- Keep a log of what each agent asked and check whether two agents got the same number
- Set up a separate workspace per client as a consultant, then hand it over
datatomcp pricing
datatomcp has a free plan with no card, a Pro plan and a custom Business plan.
| Plan | Price | What you get |
|---|---|---|
| Free | €0 | 3 datasets of up to 100,000 rows each, 1,000 queries a day, 1 workspace, 2 people, 30-day audit log, MCP server and REST API |
| Pro | €12/month (or $12) | 25 datasets of up to 2 million rows each, 50,000 queries a day, Google Sheets sync within minutes of an edit, 3 workspaces with 5 people each, 90-day audit log |
| Business | Custom | More workspaces, 20+ people, higher limits, help setting up your data and priority support |
You get an email at 80% of your daily queries; agents get a clear error at 100%, and the count resets at midnight UTC.
Pro can be cancelled any time from the billing portal; back on Free, datasets stay readable.
The optional in-app assistant runs on your own AI key (Anthropic, OpenAI, Mistral and others), so its cost is separate.
Checked 15 November 2026 · Official pricing page · Prices change, so confirm before you buy
Who datatomcp is for
- work from spreadsheets and exports and use more than one AI agent across the team
- need totals from an agent that are computed, not estimated, with the dataset version named
- want agents to have read-only access with an audit trail, and prefer data hosted in the EU
- are a consultant or agency setting up AI access to data for clients
- already have a data warehouse and a semantic layer run by a data team
- need agents to write back to the data or run arbitrary SQL: agents are read-only and limited to validated query tools
- have datasets over 2 million rows each, which is above the Pro limit (Business is a custom quote)
datatomcp pros and cons
- Exact totals come from a database rather than a model reading pasted rows
- Metric definitions need a person's approval before agents use them
- Every agent question is logged, with repeated questions compared
- Read-only for agents, no raw SQL, and results capped per call
- Data hosted in the EU, with a free plan and a public demo workspace that needs no sign-up
- A one-person company that launched in November 2026, so it is very new
- Agents can't write data or run custom SQL, which some teams will want
- Free plan limits (3 datasets, 100,000 rows each, 2 people) suit a trial more than daily team use
- Source list is fixed for now (sheets, files, PostgreSQL, MySQL, Stripe, HubSpot, Airtable); others are added on request
- ChatGPT is marked "soon" in parts of the site even though the FAQ says it can connect
datatomcp alternatives and how they compare
These tools also put AI on top of business data, each in a different way. Prices are quoted from their own pricing pages, checked 15 November 2026.
| Product | What it is | Starts at | Reviews | How it differs from datatomcp |
|---|---|---|---|---|
| datatomcp this page | Hosted data layer that lets AI agents query a team's sheets, databases and apps with exact, audited answers | Free plan; Pro €12 or $12 a month | No reviews yet | — |
| Rows | Spreadsheet with built-in AI analysis and data integrations | Free; Plus $8/month per user | — | Rows is a spreadsheet you work in, with AI tasks inside it. datatomcp is not a spreadsheet: it gives outside agents read-only, audited access to data you already keep elsewhere. |
| MotherDuck | Cloud data warehouse built on DuckDB | Lite $0 (10 GB storage, 10 hours of compute a month); Business $250 per org a month plus usage | — | A full SQL warehouse for people who write SQL. datatomcp targets teams without a data team and adds approved definitions and an agent audit log on top. |
| Hex | Data workspace with notebooks, apps and a built-in AI agent | Community plan free; Professional $36 per editor a month | — | Hex is a full analytics workspace for people who build notebooks and apps. datatomcp is narrower: a read-only source that outside agents query, aimed at teams without analysts. |
Teams with an existing database can also point an agent at a general-purpose PostgreSQL or MySQL MCP server, but then they lose the shared definitions, linking and audit log that datatomcp adds.
datatomcp reviews
Write a review
Thanks — share your take?
Frequently asked questions
What is datatomcp?
A hosted service that connects a team's sheets, files, databases and apps and lets AI agents query them through an MCP server or REST API. A database computes the answers, approved definitions say what each number means, and every agent question is logged.
How much does datatomcp cost?
There is a free plan (3 datasets, 100,000 rows each, 2 people) with no card needed. Pro costs €12 or $12 a month for 25 datasets of up to 2 million rows and 5 people per workspace. Business is priced on request.
Which AI agents work with datatomcp?
Any client that supports remote MCP servers, including Claude, Claude Code and Cursor; the site also lists Codex, Gemini CLI, n8n, Make and Zapier. Tools without MCP support can use the REST API with an API key.
Can agents change my data or run SQL?
No. According to the site, agents get read-only, validated query tools, can't run raw SQL, and get at most 1,000 rows per call.
Where is my data stored?
In the European Union, according to datatomcp. It says data is never used for training and is deleted at once when you delete it, and from backups within 14 days.
Sources (7), checked 15 November 2026
- datatomcp official site and pricing
- datatomcp chart library page
- datatomcp live demo workspace
- Show HN post, 12 November 2026
- Rows pricing
- MotherDuck pricing
- Hex pricing
We list datatomcp from public sources and are not affiliated with it. Spotted something wrong? Tell us.