Rootline is now available as an official plugin in ChatGPT and Codex, bringing payments, accounts, balances and reports directly into the AI tools our partners already use. What is now a live plugin started as an experiment during a company hackathon.
Two hours is enough time to sit through a meeting and think about what you might build. At our hackathon, it was enough to get a first version working. The idea was straightforward: connect Rootline to an AI assistant and let merchants ask about their payments. But how do you take that demo and turn it into an officially listed plugin that partners can use in their daily work? The decisions, testing and review that followed made it a much larger project than those first two hours suggested.
Rootline in the tools your team already uses
More and more work is happening inside the chat window. For years, checking a payment or a balance meant opening a dashboard and finding your way to the right page. We see conversations becoming the next standard for much of that daily checking and searching. If you can ask about your payments in the tool you already have open, that quickly becomes a natural way to work.
Developers use AI assistants while building integrations. Operators use them to investigate issues and answer merchant support questions. Founders want a quick overview of how things are going. For teams already using Claude or ChatGPT, it made sense to make Rootline available there too.
Our dashboard gives users a clear view of their payments and accounts. The plugin brings that information into the AI conversation. If a merchant asks your support team about a failed payment, you can look it up, check what happened and ask whether other payments for that merchant failed during the same period.
You could also ask how much a merchant processed last week or which payment methods its customers used. For finance teams, it might start with “How much is pending settlement?” and lead to opening the relevant report.
You might need the details of a single payment, or you might want to investigate further. Either way, you can keep asking questions in the same conversation, using information retrieved from Rootline.
So how does the assistant know where to find that information?
How the connection works
The connection runs on MCP, short for Model Context Protocol. It gives AI assistants a standard way to work with external tools and data. For Rootline, we built tools to retrieve payments, accounts, balances, reports and other records. Each tool has a description explaining what it does and what information it needs.
You connect by signing in with your existing Rootline dashboard login. Each request follows your account’s permissions, so the assistant can retrieve information from the merchant accounts you already have access to. Everyone using the plugin therefore needs their own Rootline account.
When someone asks about a particular payment, the assistant can use its payment ID to retrieve the details from Rootline. A broader question may involve several tools. To answer “How did this merchant do last week?”, for example, it first finds the merchant’s account and then uses that account to request activity for the relevant period. The tool descriptions guide these steps, helping the assistant choose which tools to use and what information to pass between them.
But to give the right answer, the assistant first needs to understand what the user is asking. A question like “Show me my payments” sounds simple, but for a platform managing several merchants, it needs a little more context.

Behind the answers
“My payments” could mean the platform’s own activity or the payments of a particular merchant. Does the user want all payment attempts, or only successful payments? A request like “Show me last week’s payments” leaves both questions open. The assistant needs enough context to choose the right account and filters, or ask a short follow-up question.
Platform payments add another detail to consider. If a shopper pays €100 and that amount is divided between a merchant and the platform, the original payment and the merchant’s share tell different stories. An answer about the shopper’s payment should show the full amount, while a question about the merchant’s share should focus on the relevant split. Counting both as separate purchases would give the wrong total.
These examples only cover payments. Questions about accounts, balances, reports and onboarding bring their own details.
There is also no way to predict exactly what someone will ask or how they will phrase it. Two people might ask the same thing in different ways, while a small change in wording can call for a different answer. Testing therefore goes further than checking whether a tool returns data. It also means checking whether the assistant understood the question and used that data correctly.
To see how the assistant handled that variety, hundreds of agents ran through different questions and conversations, alongside testing with our own team and merchants. The tests included follow-up questions, switches between accounts and deliberate attempts to trip the assistant up with unusual wording, missing details and requests it should not fulfil. Each test looked at the tools it called, the information it requested and whether its answer matched the question.
Making Rootline easier to connect
With the plugin tested, the next step was making it easy for our partners to find and connect. Compatible AI assistants can connect to Rootline through a custom setup, using the server address provided in our documentation. That gives partners a way to use it directly, but requires some configuration and depends on which custom connections their AI account or workspace allows.
An official listing makes getting started more straightforward. Partners can search for “Rootline” in the ChatGPT directory, select the plugin and sign in with their Rootline account. The connection is already configured, so they can start using it without entering a server address or setting up a custom plugin.
Getting listed also meant passing OpenAI’s review for publication. That process covered how the plugin works, how users sign in and how its tools handle requests and return information. After five weeks, Rootline was approved for the directory shared by ChatGPT and Codex. Our partners could now discover and connect it directly within the tools they already use.

From the demo to your working day
The first two hours showed that the idea could work. The development, testing and review that followed turned it into a plugin our partners can now use with their own Rootline accounts.What comes next will be shaped by the questions our partners ask and the feedback they share.
You can find Rootline in ChatGPT and Codex, sign in and start with a question about your payments. Our setup guide also explains how to connect through Claude and other compatible AI assistants.

