For agents

Teach your assistant to read paperwork

Models are good at reading one document and bad at reading four hundred consistently. Datanem does the second part: agree a set of columns once, then every document lands in the same shape. Connect it over MCP and your assistant can run the whole thing without you uploading anything.

claude mcp add --transport http datanem https://datanem.com/mcp

There is no API key. Signing in happens in your browser the first time the agent uses it, so nothing secret ever sits in a config file.

What it can do once it is connected

• Read a folder of paperwork

  Point it at a stack of invoices, receipts, statements or delivery notes and it sends them off in one job. PDFs, Word files and photographs of pages alike, including scans with no selectable text.

• Agree the columns first

  Datanem proposes a column design from the documents themselves and waits for it to be approved before reading the batch. The agent can read that design, and so can you.

• Get a database back, not a wall of text

  The result is a table your agent can query, or a ready-to-run SQL database with the line items in their own table and the suppliers in theirs.

• Reuse a design across batches

  Saved designs are listed as well, so next month's pile goes through the same columns without anybody agreeing them again.

Why a person still has to agree the columns

An agent reading a stack of invoices unattended is how you end up paying the wrong supplier. Datanem proposes a design from a sample and stops. Nothing is read in bulk until that design is approved, which means the shape of the output is decided once, by a human, rather than guessed afresh on every document.

What it costs

The MCP server is on every plan, including the free one, and billing is per document however many pages it runs to. Documents an agent submits come off the same allowance as the ones you upload yourself, so there is nothing separate to buy and nothing that can quietly run up a bill beyond what you have.

Where your documents go

Files and extracted data are held in the EU and are never used to train a model. An agent connects as you, or as the owner of your team, so it reads and writes exactly the documents the web app shows you and nothing else. The endpoint is https://datanem.com/mcp, and you can revoke its access from your account at any time.