alphabrand exposes its Data agent as a remote MCP server. Any assistant that speaks MCP over HTTP can connect to it: Claude, ChatGPT, Microsoft Copilot, or your own client. You sign in once with your alphabrand account, and the assistant can list your workspaces and ask data questions in them.
Endpoint
https://api.alphabrand.ai/api/mcp| What | URL |
|---|---|
| MCP endpoint (Streamable HTTP) | POST https://api.alphabrand.ai/api/mcp |
| Protected resource metadata | GET https://api.alphabrand.ai/.well-known/oauth-protected-resource/api/mcp |
| Authorization server metadata | GET https://api.alphabrand.ai/.well-known/oauth-authorization-server |
Authentication is OAuth 2.0 with dynamic client discovery. A compliant host finds the authorization server from the metadata above, registers itself, and sends you to the alphabrand sign-in. There is no API key, client id or secret to paste.
What you need
- Market data enabled for your organization and assigned to your workspace, the same requirement as the
@Dataagent in the app. - At least one brand in the workspace, and the right to create projects in it.
- An assistant that supports remote MCP servers over HTTP with OAuth. The next pages cover Claude, ChatGPT and Microsoft Copilot.
What the server exposes
Three tools. The full input and output shapes are on the tools reference.
| Tool | What it does |
|---|---|
list_workspaces | Lists the workspaces you can query and the datasets each one gives access to. Call it first. |
describe_workspace | How to ask in one workspace: each dataset's coverage (geography, time grain, period) and the house definitions the agent applies, in your company's own vocabulary. Call it once before the first question. |
ask_data_question | Asks a business question in natural language. Returns the answer, its assumptions, the datasets used and a conversationId for follow-ups. |
The agent owns the methodology
The data agent applies your company's definitions, market scope and current data period on its own, the way an analyst of your company would: what counts as an innovation, the default market scope, the reference period, how share and growth are read. An assistant does not need to re-derive any of that from a past report.
describe_workspace tells the assistant which terms are defined so it can name them in its questions ("top 20 spirits innovations", "share of motherbrand", "key markets") instead of defining its own. Every answer lists the assumptions it applied; that list is authoritative on scope, period and definitions, and is meant to be quoted next to the figure. The SQL, column mappings and formulas stay on the server.
What happens in alphabrand
Each question asked from outside creates a project in your workspace, titled from the question and marked with an MCP badge. Follow-ups that pass the conversationId continue the same Data conversation, so the agent keeps its earlier SQL, assumptions and rows. Open the project in alphabrand to see the charts; the assistant receives the text answer. Corrections you make from outside, "always exclude travel retail", are saved to the project memory like any other.
Only the Data agent is available this way. Consumer Insights, Insights and the document agents are not.
Timing
A first answer takes about half a minute, a follow-up about a minute. The server sends MCP progress notifications when the host passes a progressToken, which keeps the call alive on hosts with idle timeouts. Hosts with a hard tool-call timeout under a minute may still cut long answers off.