What is an MCP?

You've probably seen the term "MCP" a dozen times this month without anyone stopping to explain it. It stands for Model Context Protocol, and it's essentially an API for AI agents — a way of telling an agent what data is available, what each part of it does, and how to ask for the right thing.

If APIs aren't your world, think of it as a set of instructions that lets an AI assistant find and use data on its own — so instead of you digging up the numbers and handing them over, the assistant goes and gets them itself.

Why do MCPs matter?

Every day in our personal lives, looking something up on the internet used to mean opening Google, searching, clicking into one page and then another, and reading three or four of them before you'd pieced together an answer. Now you just ask ChatGPT or Copilot the question directly and it does all of that for you — the searching, the reading, the comparing — and hands back a single answer.

An MCP allows you to have the same experience with the data that lives in your enterprise tools. Think of it like an API for your AI agents. Without one, an AI assistant can't see your traffic data, which leaves you doing the fetching yourself: exporting files, formatting them, pasting them into the chat window. With an MCP, the assistant reaches the data directly, so a single question gets you a single answer and the agent does the legwork you used to do by hand.

The problem it solves

Agencies are already leaning on ChatGPT, Copilot, and Claude to speed up their analysis and reporting, but getting their own data into those tools usually means exporting spreadsheets and dropping them in by hand. That creates more work rather than less for users, and it only goes so far — plenty of vendors don't make their data easy to get out at all, with no clean export, no open API, and nothing an agent can actually reach.

We've taken the opposite approach. At Flow Labs we're making the data in the Flow platform as accessible as we can through FlowMCP, so your agent can reach it directly and securely, without you playing middleman.

What agencies are already doing with FlowMCP

A few of the things customers are using it for today, each of which we'll go deeper on in posts to come.

‍

Measuring the impact of a retiming program. Ask for a before-and-after across the signals you changed and see the delay and travel time you actually saved, project by project, in minutes rather than weeks.

Prioritizing signals and corridors for retiming. Have your agent rank the network by where retiming will do the most good, so your budget lands on the intersections that need it rather than the ones that complained loudest.

Collecting data, including turning movement counts. Pull a turning movement count for any intersection, on demand, without sending a crew into the field.

Analyzing performance for a signal or corridor. Ask why a corridor backs up every afternoon and get the answer down to the phase and movement, without spelunking through a dashboard to find it.

Weeks of work, done in fifteen minutes — your way

Each of these reports used to be a project in itself: pulling the data, formatting it, rewriting it after feedback, then formatting it all over again — work that could eat days or weeks. FlowMCP turns it into a less than fifteen-minute task, delivered the way you wanted it the first time.

And it's delivered on your terms. Because FlowMCP works on a bring-your-own-agent model, you're not boxed into how we think the work should be done. You can build your own skills, playbooks, and templates so the agent produces the analysis and the report in your format and to your spec — your process and your output, with our data underneath it.

If you're already using AI, this is your next step

If you're already using Copilot or ChatGPT to speed up your analysis and reporting, you've felt the first half of this shift. FlowMCP is the second half — when your AI stops waiting on you for the data and starts working straight from the source.

The reports keep getting longer while the time to act on them keeps getting shorter. FlowMCP tips that balance back: less of the week spent assembling information, and more of it spent making the calls that keep people moving. Engineers doing engineering again.

If you're ready to see what your own agent can do with your data, talk to our team about FlowMCP at contact@flowlabs.ai.