Skip to content
Last updated


Purpose

Connect a Search MCP to a mentor so it can search your platform for courses, programs, and mentors—and return grounded, recommendation-ready results without hallucinations.


Overview

The Search MCP uses an MCP server (in this demo, ibl.ai’s own search MCP) to power a mentor that acts as a search assistant. It can:

  • Search the course catalog (courses/programs).
  • Find and recommend mentors.
  • Ask follow-up filters (subject, level, format, language).
  • Enforce guardrails via the system prompt.

Prerequisites

  • A mentor with a Search Assistant system prompt.
  • MCP Tool enabled for the mentor.
  • An API key for the Search MCP server.

Setup Steps

1) Prepare the Mentor

  • Open the mentor you want to use for search.
  • Set the system prompt to a search-focused assistant (catalog + mentor search, recommendations, guardrails).

2) Enable the MCP Tool

  • Go to the mentor’s Tools tab.
  • Ensure MCP is enabled.

3) Add the Search MCP Connector

  • Open the MCP tab.
  • Add (or edit) a connector with:
    • Connector name
    • Server location (Search MCP server URL)
    • Description
      • Acts like instructions for the connection (what the mentor can pull and how to respond).
    • Transport: Streamable HTTP
    • Authentication: API Key
      • Token type: API Key
      • Token value: paste your token
  • Save the connector.

Using Search in Chat

Search Courses

Prompt example:

“What courses are available on the platform?”

Result:

  • Returns a subset of courses from the tenant (even if the catalog is large).
  • Asks if you want to filter by subject, level, format, or language.

Search Mentors

Prompt example:

“What mentors can help me become a better student?”

Result:

  • Lists relevant mentors (e.g., study tips, quizzes, Socratic support).
  • Offers to filter further or show more results (based on system-prompt limits).

Result

With Search MCP connected, mentors can reliably discover courses and mentors across your platform, guide users with filters and recommendations, and return accurate, grounded results from your own data.