Connect MCP with Gemini

Configure the B6AI MCP server so Gemini can build and manage your bots in natural language.

5 min readLast updated 19 August 2026

Overview

B6AI ships a hosted Model Context Protocol server, `b6ai-bot-builder`, that lets an AI client build and manage your bots (intents, utterances, knowledge and flows) from natural-language instructions. Once Gemini is connected, you can ask it to "create a support bot that handles order status and refunds" and it authors a complete bot in your workspace.

The server speaks MCP over HTTP and authenticates every request with your workspace identity, so Gemini only ever sees what your role is allowed to see.

Before you start

You need an active B6AI workspace and permission to build bots (Enterprise Admin or Agent Admin). The Gemini CLI reads MCP servers from its settings file.

Prerequisites

A B6AI workspace with the MCP server enabled on your plan.
The Gemini application installed and updated to a version that supports remote MCP servers.
Your B6AI MCP endpoint (https://mcp.b6ai.ai/mcp) and a way to sign in, the server authorizes you through B6AI’s identity provider (Keycloak / OAuth 2.0).
The OAuth client ID `b6ai-mcp`. B6AI does not allow clients to register themselves, so Gemini cannot obtain this automatically.

Get your B6AI MCP endpoint

The MCP endpoint is workspace-specific. Copy it from your workspace, or ask your administrator for it. Authentication uses OAuth 2.0 against B6AI’s identity provider, the client opens a sign-in window the first time it connects, so you do not paste long-lived secrets into config files where the client supports OAuth.

text

https://mcp.b6ai.ai/mcp

Use the B6AI OAuth client ID

Most MCP clients try to register themselves with the authorization server the first time they connect (dynamic client registration). B6AI switches that off on purpose, because anything that can reach the identity provider could otherwise create its own OAuth client. Instead, B6AI publishes one pre-approved client that every user shares.

When Gemini asks for an OAuth client ID, enter the value below. Leave it blank and registration fails before you ever reach the sign-in page.

text

b6ai-mcp

This is not a secret

The client ID identifies the application, not you. It is a public value, safe to share with your team and to paste into a client’s settings. Your own credentials are only ever entered on the B6AI sign-in page.

Add the server to Gemini settings.json

Add the B6AI MCP server to Gemini using the configuration below, then restart the client so it loads the new server.

json

{
  "mcpServers": {
    "b6ai": {
      "httpUrl": "https://mcp.b6ai.ai/mcp"
    }
  }
}

Keep tokens safe

If your client uses a static bearer token instead of interactive OAuth, treat it like a password: never commit it to source control, and rotate it if it is exposed.

Verify the connection

Run the Gemini CLI and use /mcp to confirm the b6ai server and its tools are listed.

Open the tools / connectors list in Gemini and confirm `b6ai-bot-builder` appears as connected.
Ask Gemini to list your bots, or to describe what it can do with B6AI.
Ask it to create a simple test bot, then open Bot Builder in B6AI to confirm the bot was created.

Connected

Gemini can now build and manage bots in your B6AI workspace with your permissions.

Troubleshooting

Symptom	Likely cause	Resolution
Could not register with B6AI’s sign-in service	No OAuth client ID supplied; B6AI does not allow self-registration	Enter `b6ai-mcp` as the OAuth client ID in Gemini, then connect again.
Server not listed	Config not loaded	Check the config path and fully restart Gemini.
Sign-in loop / 401	Auth not completed or token expired	Re-run the sign-in flow, or regenerate the token and update the header.
No bots returned	Account has no bots or lacks permission	Confirm your B6AI role can build bots and that the workspace has at least one bot.
Tool calls rejected	Endpoint or transport mismatch	Confirm the URL ends in `/mcp` and matches the transport Gemini expects.

Was this page helpful?

Connect MCP with Codex

Creating Your First Bot