1. Understanding the Model Context Protocol

What is the Model Context Protocol (MCP)?

MCP is an open standard that lets an AI model treat external services as callable tools. Instead of the AI being limited to generating text, MCP gives it a structured way to act: read your email, create a task in a project tracker, query a CRM, push a row to a spreadsheet — any service that exposes an MCP server becomes an extension of the AI's capabilities.

In OUPI, the MCP Hub is the centralized control plane for these connections. You authenticate once with a service (typically via OAuth), define the scopes you're comfortable granting, and every AI-powered surface in the platform — ChatPro, OUPI Studio, the Assistant, workflow steps, RAG imports, Deep Research — can immediately leverage that service. No per-tool wiring, no duplicated credentials. Connect once, use everywhere.

Architecture at a glance: Catalog → Connection → Tools → Execution

1. Server catalog — A browsable directory of available MCP services. Each entry describes what the service offers before you connect.
2. Connection — Created when you authorize a service. It stores the OAuth grant and the scopes you approved. A connection is platform-wide: every OUPI feature that can invoke tools sees it.
3. Tool list — Each connection exposes a set of tools. Open a connection to inspect them; this is exactly what the AI model sees when deciding whether to call the service.
4. Execution log — Every tool call is recorded with the tool name, timestamp, originating feature, and outcome. This gives you full auditability over what the AI did on your behalf.

Understanding this pipeline is key: the catalog is discovery, the connection is authorization, tools are capability, and the log is accountability.

Security model and cost implications

Connections rely on the service's own OAuth authorization flow — OUPI never stores raw passwords. The AI can only reach what the granted scopes allow, and every invocation is logged in the execution history, so you can audit actions at any time. For sensitive operations, calls can be gated by approval workflows.

On the cost side, MCP tool calls are not billed separately. They are folded into the credits of the AI operation that triggers them — whether that's a ChatPro conversation, an agent run in Studio, or a workflow step. This means you can reason about cost at the operation level without tracking individual tool invocations.