Hard-to-reach data for every agent

Why this conversation

Agents are only as useful as the data they can get at, and some of the most useful data, from a 3D model of a building to a live sensor feed, is the hardest to reach. Getting at that kind of data is Safe Software’s business. Matt Meeboer, from the company, talked to the host on 23 October 2026, the second day of AGNTCon + MCPCon North America in San Jose, about how it now hands that data to agents through MCP, what customers do with it, and what comes next. The rest of the day’s conversations are in the article on day two.

From geospatial to almost anything

Matt describes Safe Software in one line: “We’re a data integration and orchestration platform.” The company has existed for over 30 years. It began in a niche, the GIS space, which the host spelled out as geospatial, and has grown over those decades to integrate with almost any system. Even the name comes from that start: back in the 90s there was a geospatial format called SAFE, and a SAFE file.

What brought the team to the conference was MCP, mostly. In his words, “now we’re connecting any data source to any agent using MCPs.” He points at the data that is hard to reach, for instance the 3D model of a building like the one they were standing in. The question is how to put that in the hands of an agent. Safe Software already has the connectors, and the means to validate such data, transform it and turn it into a structured format an agent understands.

MCP servers without code

The new piece arrived in the 26.2 release of FME Flow, which came out earlier this year. With it, a user can create one or more MCP servers, each with MCP tools, without writing code, and use them to deliver that hard-to-reach data.

The host asked whether this connects to embodied AI, a term he credited to Fei-Fei Li, about agents that move through a 3D and 4D world. Matt had not heard it. For Safe Software, he says, the work is about building tools, as a software vendor, for customers who handle every kind of dataset, spatial and otherwise, so they can pass that data on to agents.

The pipelines people build in FME are no-code and easy to move around. They are stored in a file format that can be published to the company’s web server, where a pipeline becomes the code behind an MCP tool, or behind a no-code automation for a web application.

Traffic sensors and buses

Asked for the biggest use case, Matt says FME is used in too many ways to pick one, and offers two recent examples, recent because all of this is still new. A company in the utility space, whose name escaped him, monitors traffic. It ingests vehicle traffic data from real-time sensor feeds, which produce a great deal of it, and now gives that data to agents through MCP servers on FME Flow, so that semantic questions can be asked of it.

Another customer works in transit. Its buses and other vehicles report their position every 10 seconds, which the host found a lot. The customer wants to transform that stream and thin it out before handing it to any agent, and that step is where FME fits.

Growing with the protocol

Safe Software adopted MCP early. After so many years of connecting to so many kinds of data, Matt sees MCP as the gateway that lets agents talk to everything the company already integrates with. The protocol is young and still being developed, with asynchronous capabilities among the things on the way, and the company intends to grow along with it.

The other big topic at the conference, he says, was security and control. On the platform you can already see exactly what a tool call has done. The plan is to improve that lineage and make it easier to discover, so a user can see graphically what each call did. The company’s user conference is in London in March of next year. To get in touch, he points to the website, fme.safe.com, which has ways to reach the team; they are happy to set up a demo and learn about your use cases.

When it comes out

The episode is edited and in the release order. Like every episode it goes out on YouTube first, then on Apple Podcasts, Spotify and Amazon Music.