We translated a marketing agency’s brand-discovery methodology into an application that runs inside an AI assistant. Through the Model Context Protocol (MCP), guided interviews, audience research, owner approvals, and staged generation become one connected workflow. The result is a structured brand reference built from the owner’s knowledge and attributable evidence.

01

Encode the methodology, preserve the conversation

The agency’s discovery process became an authored interview flow: what to ask, which answers need clarification, when to gather evidence, and where the owner needs to make a decision. The assistant works through that flow in small conversational steps. It preserves the owner’s language while the application keeps the process consistent, reviewable, and resumable.

02

A conversation with a reliable backbone

The assistant handles the dialogue while the server owns interview state, validates writes, and determines the next action. Questions arrive in small, related groups. The owner reviews proposed wording before a confirmed batch is saved. Progress is persisted so the experience can resume with the correct context.

03

Evidence stays attached to the answer

Owner statements, prior material, and research are distinct sources. The workflow asks for permission before using historical material and distinguishes direct audience evidence from a brand owner’s interpretation. Thin or conflicting answers lead to clarification instead of a fabricated conclusion.

04

From discovery to an artifact

Confirmed inputs feed an asynchronous build that develops the brand reference in stages. Embedded milestone cards present the resulting material while the conversation remains the place for review and decisions. The application coordinates progress, recovery, and result delivery across the assistant and backend.

05

A complete application around the interview

The implementation brings together the MCP server, persistent interview state, research and approval checkpoints, an asynchronous generation pipeline, and embedded result cards. Each part has a defined responsibility: the assistant conducts the conversation, the server enforces the workflow, and the owner controls the decisions. That gives the agency a repeatable way to move from discovery to a structured, reviewable brand reference.