What is MCP, and what does it mean for MMM?

An AI assistant on its own knows nothing about your marketing data and cannot run a model. MCP is the protocol that connects it to systems that do. The system publishes a set of tools (run a forecast, list the channels, create a scenario, query the spend) and the assistant calls them in a conversation, in natural language, with the computation done on the system's side. Any assistant that speaks MCP can connect to any MCP server.

Odins is an MCP server. A customer connects their own Claude, ChatGPT or Copilot to it and asks the questions they would otherwise click through: how much do we need to spend next month to reach this target with 80% certainty; what would Odins have done differently last year, and how would it have turned out; which weeks should we lean into before Christmas. Every answer is computed by the Odins platform, the same scenarios and Monte Carlo runs the platform itself would produce, and returned with an explanation of what the numbers mean. The assistant does not estimate anything; it asks Odins. A scenario created in the chat sits in the platform with a link, so anything the assistant says can be checked.

The model's setup is exposed through the same door: how the channels are defined, how the priors were set, what the control variables are. A CFO gets a direct answer; an analyst can dig as deep as they want.

Since the launch, the large majority of platform use across Odins customers goes through the MCP, from the customer's own assistant rather than an Odins chatbot, and more people inside each customer use it than ever used the interface. Nobody has to learn a new tool. It is also the API: an agent connected to the MCP can forecast, create scenarios, write and pull data, so a data team that wants numbers on its own dashboard points its agent at the MCP, and the interpretation of each value is encoded in the server, so the agent knows what the results mean.

Before a model is built, the same connection lets a team talk to its history and media plans, and ask about the raw data at ad level.

• Scenario planning: Forecasting what a budget would return before committing it, and comparing several futures side by side.
• Structured marketing data: Every source, one dataset, your hierarchy, your warehouse. The foundation the model sits on, and a product on its own.
• Forecast verification: Forecast first, compare with actuals afterwards. The only honest way to know whether a model is right.

Go one level deeper: Odins MCP. Marketing mix modelling from ChatGPT or Claude. Or go back to all terms in the glossary.