Build a customer-facing assistant with Inkeep Unified Search

Step-by-step tutorial to build an AI assistant that answers your customers' questions, grounded in your knowledge base via Inkeep Unified Search.

Customer Assistant sub agent connected to Inkeep Unified Search MCP on the Visual Builder canvas

Overview

In this tutorial, you'll build a customer-facing AI assistant in the Inkeep Visual Builder. When a customer asks a question about your product, the agent will:

Search your knowledge base via Inkeep Unified Search
Answer in plain language, grounded in what your docs actually say

Prerequisites

Access to Inkeep Enterprise — schedule a demo if you don't have it yet
An Inkeep account with Inkeep Unified Search configured for your product (knowledge base, docs, sources connected)

Create the agent

In the Visual Builder, go to the Agents tab in the left sidebar and click Create agent.

Click the Default Sub Agent on the canvas to configure it. Fill in:

Name: Customer assistant
Prompt: Paste the prompt below
Base model: Leave the inherited default (e.g., Claude Sonnet 4.5) unless you have a specific reason to change it.

Prompt:

You are a customer assistant for [Your Product]. Answer customer questions using the official documentation.

How to answer:
1. Always search the knowledge base first to look up relevant docs before answering.
2. If the docs don't cover it, say so plainly. Don't invent facts.

Tone: Direct, neutral, no fluff. Speak in the first person on behalf of the company.

Tip

Replace [Your Product] with the name of the product the assistant represents.

Connect Inkeep Unified Search via MCP

Your agent now needs the ability to actually search your knowledge base. Inkeep Unified Search is exposed as an MCP server you can connect to — once wired up, the agent can call search-knowledge-base to ground its answers in your real content.

You'll do this in three parts:

Get an API key from your Inkeep Portal
Save the API key as a Credential in the Visual Builder
Register the Unified Search MCP server in the Visual Builder, then attach it to your agent

Get an API key from your Inkeep Portal

Click Assistants in the left sidebar, then Create assistant → API (Server-side).

Note

The Portal also has an MCP option in the same dropdown — skip it. That option spins up a new MCP server; we're connecting to Inkeep's existing Unified Search MCP, which just needs an API key for auth. API (Server-side) is what generates that key.

Name the assistant something memorable (e.g., customer-assistant) — make a note of the name, you'll need it to construct the MCP URL in a later step. Then copy the API key shown after creation.

Save the API key as a Credential

Back in the Visual Builder, click Credentials in the left sidebar, then New Credential.

Choose the Bearer authentication option and fill in:

Name: A memorable name (e.g., inkeep-unified-search)
API key: Paste the API key you copied from the Portal
Headers (optional): Leave blank

Bearer credential form in the Visual Builder

Click Create Credential.

Register the MCP server and attach it to your agent

Click MCP Servers in the left sidebar, then New MCP Server → Custom Server.

Fill in:

Name: Inkeep Unified Search
URL: https://mcp.cloud.inkeep.com/<your-assistant-name>/mcp — replace <your-assistant-name> with the name you gave the assistant in the Portal (e.g., customer-assistant → https://mcp.cloud.inkeep.com/customer-assistant/mcp)
Credential: Select the credential you just created

Click Create.

Go back to your agent canvas: Agents → Customer assistant.

Drag and drop an MCP block from the top left toolbar onto the canvas. Select Inkeep Unified Search and connect it to your sub agent.

Click Save Changes in the top right corner.

Test your agent

Click Try it in the top right corner to open the chat interface.

Ask a product question your docs should be able to answer — for example, "How do I get started?" or "What are the main features?"

When working correctly, the agent will search your knowledge base and answer grounded in what it finds. If the docs don't cover the question, it will say so rather than guess.

Next steps

You now have a working customer assistant grounded in your knowledge base.

Add citations — let customers verify each answer at the source with inline footnotes and source cards.
Add human escalation — when the agent can't help, hand off to a human via a support ticket with full conversation context.

Build a customer-facing assistant with Inkeep Unified Search

On this page