Skip to main content

What is the Predictive Index MCP connector?

This article applies to client-side IT administrators and technical sponsors responsible for enabling PI's MCP connector during the current Pre Beta phase.


The Predictive Index MCP connector uses the Model Context Protocol (MCP) — an open standard — to bring Obi, PI’s behavioral intelligence product, into the AI tools your team already uses: Claude, Microsoft Copilot, and ChatGPT. Once enabled, users can ask behavioral intelligence questions from inside their AI host and get answers grounded in PI’s validated behavioral science, with citations.

There is nothing to download or package. Enabling the connector means registering a URL in a supported AI host and having each user sign in with their own PI account. During the Pre Beta phase, access is controlled by who you enable the connector for, and by each user’s existing PI permissions.

Note: Use of the Predictive Index MCP connector requires that its users also have access to the following:

Quick facts about the PI MCP connector

Category

Details

Intended users

Any PI users, from a small group to all employees, as determined by your PI administrator

Supported hosts

Microsoft Copilot — Microsoft 365 Copilot, Copilot in Microsoft Teams

Claude — Claude.ai (web), Claude Desktop, Claude Team/Enterprise


ChatGPT — ChatGPT on the web, added as a custom app. Requires developer mode

Deployment model

Register a remote MCP connector (URL) in the host, then per-user sign-in with PI credentials.

Connector URL

Authentication

PI single sign-on via OAuth 2.1 (authorization code + PKCE)

User provisioning

Every user needs a valid PI platform account with Obi access.

Capability

Read-only. Obi can look up, explain, and generate content; it cannot send, email, invite, save, or change anything.

Support contact

Stefano Hernandez, PI Product Manager ([email protected])

How the PI MCP connector works

The connector exposes three high-level, read-only tools to your AI host. Based on what the user is doing, the host picks the right one:

  • Ask Obi — for a PI, behavioral science, person, team, hiring, or coaching question

  • Review Message for Recipient — when the user wants feedback on the wording or tone of something they are about to send to a specific person: "How will this land for them? How should I reword it?"

  • Prepare for Conversation — when the user wants to prepare for, or decide how to approach, a live conversation, meeting, or response with a specific person

Whichever tool the host calls, it sends the request to PI’s deployed Obi service, which applies PI’s science, tools, and safety pipeline and returns a validated answer with PI citations. The host then relays that answer to the user.

Some additional notes:

  • All three tools are read-only; they review and advise, but never send a message or change any data.

  • The connector is a thin, authenticated bridge. It does not ship PI’s science to the AI host, and it runs every request through PI’s own security checks server-side; the AI host is not trusted as a security boundary.

  • The connector honors each user’s existing software permissions, meaning it won’t expose what users don’t have access to in the PI software.

  • It is stateless. Each question is answered fresh and is not added to the user’s Obi chat history in the PI web app.

Additional support


Did this answer your question?