Ask our data a question: Point Topic's broadband ontology is now open to AI assistants

Most analysts we speak to have started asking questions through an AI assistant. The awkward part is what happens next. The assistant is fluent about broadband in general and knows nothing about who actually owns which network in the UK, so it fills the gap with plausible guesses. That is worse than a blank page, because a plausible guess about network ownership reads exactly like a fact.

We have opened a route around that. Point Topic's broadband ontology — the structured map of operators, networks and the relationships between them — is now available to any AI assistant that speaks the Model Context Protocol, with no account and no API key.

Visit this page for instructions on how to get set up: https://mcp.point-topic.com/

What is in it

The public endpoint serves our reference layer: the entity graph rather than the measurements. As of today that covers 1,120 organisations, of which 979 are ISPs and 767 are network operators, across 1,615 networks. Sitting on top of those are 2,119 recorded ISP-to-network selling relationships, 1,496 ownership links between operators and the networks they own, and 27 link standards describing the technologies those networks run.

The relationships are the part that is hard to assemble anywhere else. A retail brand, the network it sells over, and the operator that owns that network are three different things, and in the UK they are frequently three different companies. Wholesale arrangements, rebrands and acquisitions mean the answer to "whose fibre is this customer actually on" changes several times a year. The ontology tracks those entities with validity dates, so the graph reflects the market as it stands rather than as it stood when a press release went out.

Questions the public endpoint answers well:

• Which networks does a given ISP sell over, and who owns them?

• Which ISPs are available on a particular network?

• What technology standard does a network run on?

• How does one operator's set of networks compare with another's?

Discovery is built in. An assistant connected to the endpoint can list the available tables and inspect their columns, so it works out the shape of the data without being told.

Where the line sits

Our measurement data stays licensed. Anything counted or modelled — coverage and footprint, premises, speeds, tariffs, forecasts, take-up, subscriber numbers — is a Point Topic product and is not served on the public endpoint. A query that reaches for it gets a clear notice that the data exists but access is restricted, along with a route to ask us about it.

That was a deliberate design choice. The easy implementation returns an empty result and lets the assistant conclude the data does not exist, which would leave a model quietly telling users we have no coverage data. We would rather the boundary be visible. Licensed clients connect to the authenticated endpoint instead, and see exactly the datasets and geographies their subscription covers.

Connecting

It is a plain Streamable HTTP MCP endpoint, so any MCP-capable client can reach it, including Claude Code and Claude Desktop, Cursor, and ChatGPT. Per-platform setup notes live in the README at github.com/Point-Topic/point-topic-mcp.

We are as interested in the questions as in the answers. If you connect it and find something the ontology should know and does not, tell us — that feedback shapes what goes into the reference layer next. Get in touch at info@point-topic.com.