Lunch & Learn: MCP for Public Sector Data Silos

If you work in a government department, you already know the drill. The answer to a straightforward question is sitting across four systems, two clouds, and a spreadsheet someone emailed round in 2022. Analysts spend more time logging in, exporting, and reconciling than they do thinking. Cross-team sharing is worse. Cross-department sharing is a running joke.

The standard fix is to consolidate. Pick a central platform, map every schema, build the pipelines, migrate the data, retrain the teams. Two years later you've got half the sources onboarded, the other half pushing back, and a governance model nobody likes. The legacy systems you most needed to reach are still sitting there untouched, because getting data out of them cleanly was never realistic in the first place.

Model Context Protocol takes the other path. It's an open standard that lets AI systems talk to data sources through lightweight server wrappers, so instead of moving the data you wrap it where it lives. One interface, many back-ends. An orchestration layer queries across them at once, reconciles the conflicts, and hands back a single coherent answer with sources attached. No migration. No new platform for anyone to learn. The data stays put.

This isn't hypothetical. We've been shipping MCP layers into real public sector estates, bridging data across AWS and Azure, across legacy kit and modern platforms, without moving a single byte. We'll walk through what those deployments actually looked like, the bits that worked first time, and the unglamorous bits that took a few goes to get right.

This is a lunch-and-learn, not a demo reel. We'll spend most of the hour on concepts, trade-offs, and honest failure modes. Where MCP fits, where it doesn't, and what you need in place before it's worth a punt. Bring questions. Bring the annoying edge cases from your own estate. We'd rather talk about those than run through slides.

It's aimed at civil servants, analysts, and technical leads who recognise the silo problem and are tired of hearing that the answer is another two-year data programme. By the end you'll have a clear picture of what MCP is, how it compares to the central-platform approach you've probably already been pitched, and whether it's worth a closer look for whatever you've got on your desk.