Skip to main content
Version: v2

BAV MCP Server

Your AI models and agents can use the official BAV MCP server to access BAV data in both a simple and secure way. It uses the latest MCP (Model Control Protocol) standard to provide a set of tools that your agents can call to retrieve data from BAV.

An MCP server is a convenient way to connect your AI models and agents to external data sources, such as BAV. The server offers a set of tools that agents can trigger to get the data they need to then process and use in their tasks.

We're excited to see how you and your agents use BAV data to power your workflows. If you have questions, feedback, or requests for new MCP tools, please let us know.

Authorized Agent Environments

Please note that you may only connect to the BAV MCP server from authorized agent environments that have been vetted by WPP legal. This is to ensure the data security and compliance around our proprietary data. If you are not sure whether your environment is authorized, please reach out.

Connecting to the server​

Our remote MCP server supports both SSE (Server-Sent Events) and Streamable HTTP. To connect to the BAV MCP server, you need to use the following URL:

https://mcp.wppbav.com/sse

You also need to provide an API key as a Bearer token in the Authorization header of your requests. You can obtain an API key through the API keys section of the BAV platform.

Note: When creating your API key please make sure to select all the scopes under "Tools".

Available Tools​

The BAV MCP server provides the following tools:

  • get_pillar_data
  • get_powergrid_data
  • get_imagery_data
  • get_personality_data
  • get_usage_data
  • get_preferences_data
  • get_recommendation_data

Each tool accepts a set of parameters that your agent will try to extract from the user input. This includes:

  • Brands
  • Countries (with a note to prefer using ISO 3166-1 alpha-2 codes)
  • Years
  • Audiences

Behind the scenes the BAV systems will automatically handle fallbacks should a parameter not be provided. However, the user must always provide either a brand, a country or a year. Otherwise, the agent will receive an error back.

Unstable Responses

As the MCP server is still in development the format of the responses is subject to change. We will not be reaching out to notify you of these changes until the server is stable.

We highly recommend using a reasoning model that not only can handle the tool calling effectively, but can also reason about the data returned by the tools. The following reasoning models have been tested and are known to work well:

  • Anthropic's Claude 4
  • Anthropic's Claude 3.7 Sonnet Thinking
  • Google's Gemini 2.5 Pro

Reasoning models are able to better understand the user requests, extract the necessary parameters, and interpret the data returned by the tools. This leads to more accurate and relevant responses. They are also able to better determine multiple tools that need to be called in sequence to answer a user request.