MCP Is the New "Import from Excel"

A few days ago I wrote that vibe coding is the new Excel. Today I want to come back to Excel from a different angle: not as a tool for building applications, but as the way users feed data into them.
The Data Entry Screen Wasn't Enough
For years we designed good interfaces for entering information: clear forms, inline validation, search, sensible defaults, keyboard shortcuts. And still, in almost every project, the same request showed up after a while: "Can we import from Excel?"
It wasn't that the interface was badly built. It was that the user had hundreds of records to enter, or received them from another system, from a supplier, from a spreadsheet they were already using. With an import they could:
- Load a lot at once. What took days on screen took minutes with a file.
- Prepare it beforehand. They worked on the data at their own pace, in a tool they knew well: sorting, fixing, filling in gaps and cross-checking with other sources.
- Review it all together. They saw the whole set before sending it, instead of one record at a time.
The data entry screen stayed useful for day-to-day work and one-off cases. The import handled the volume.
The Same Thing, Now with Agents
Today we're seeing something very similar with MCP. The Model Context Protocol is an open standard that lets an AI agent connect to an application and use its functions: query data, create records, run actions.
Adding MCP support to an existing application plays the same role the Excel import used to play: it gives users a fast lane for getting data in, except that now it's their agent doing the loading. And the benefits look a lot alike:
- Faster. The agent creates in seconds what takes a person form after form.
- With the information already prepared. Just like with the spreadsheet, the user does the groundwork, now with help: the agent reads an email, a PDF, meeting notes or an export from another system, organizes the data, fills in what's missing, and only then loads it.
- Orchestrated. The agent can chain steps and systems: look up a customer, create it if it doesn't exist, enter the order and attach the document, in that order.
- Correct. It doesn't copy and paste between screens: it uses the application's operations, with their data types and rules, and if something doesn't comply it gets an error it can understand and fix.
Like the Import: with Rules
There is an important nuance, and it's the same one the Excel import had. A good import was never "paste rows into the database". It validated every row, rejected what didn't comply, respected who could load what, and often showed a preview before confirming.
An MCP layer has to work the same way:
- Permissions. The agent acts on behalf of a user and can't do anything that user can't do from the interface.
- Validation. The business rules are the same, whether the data comes in through a form, a spreadsheet or an agent.
- Confirmation for what matters. Saving a draft is not the same as approving a payment, deactivating a customer or sending something to a third party. For actions with consequences, the application asks a person to confirm.
- Traceability. There is a record of what the agent did, when, and on whose behalf.
We sum it up with a phrase we repeat often: AI proposes, people decide.
UI-less Applications, and MCP Layers on What Already Exists
This opens two interesting doors.
The first is UI-less applications, or nearly so. If a good part of data entry and queries goes through agents, some applications can be born with very little screen: just enough to configure, review and approve.
The second seems closer at hand for many companies: new developments that are, essentially, an MCP layer on top of existing or legacy applications. Systems that work, that hold years of business rules, and that nobody wants to rewrite, but whose interface has become slow or awkward for the way people work today. Instead of rebuilding everything, you can build a layer that exposes their operations to agents and respects the permissions and validations the system already has. It's a way to modernize how a system is used without touching its core.
What We're Working On
At Flowing Code we're working on adding MCP support to our Geni solutions. And we've already done it on an internal social media post planning product, which we're considering releasing to the public.
If you have an application, new or years old, where entering information has become a bottleneck, let's talk. We can help you open it up to your users' agents, with the same rules as always.
Join the conversation!