Skip to main content

Connect an MCP server

Cubeo AI lets you bring tools from any Model Context Protocol (MCP) server straight into your AI Agent. This guide walks through connecting a server in under a minute.

Connect once, use everywhere​

You only connect an MCP server once per team. After that, every AI Agent in your team can add the server's tools from its Tools tab — there's no need to reconnect for each agent. The tools you see fall into two groups:

  • Curated by Cubeo admins — MCP servers admins have set up for everyone on the platform. You don't need credentials; they're ready to use.
  • Connected by your team — MCP servers you (or someone else on your team) connected. They're scoped to your team only.

Both groups appear together in the AI Agent's Tools catalog with an MCP badge so you can tell them apart from native integrations.

What you need​

  • The MCP server URL (something like https://mcp.example.com/mcp).
  • Optionally, credentials for the server. Cubeo auto-detects what authentication the server expects, so you only fill in the fields the server actually needs.

Where you connect from​

There are two entry points — both open the same wizard:

  • AI Agent → Tools tab → Connect MCP. The server's tools are added to this agent right away, and the connection becomes available for your other agents too.
  • Team → Integrations → Connect MCP server. Use this to set up a connection without targeting a specific agent first. The tools then show up in every AI Agent's Tools tab, ready to be added with one click.

Step 1 — Paste the URL​

Paste the server URL (for example https://mcp.tavily.com/mcp/?tavilyApiKey=...) and click Continue.

Cubeo checks the server, looks at how it expects to be authenticated, and moves you to the next step. If the URL is not actually an MCP server, Cubeo tells you so and nothing is saved — just fix the URL and continue.

Step 2 — Confirm the authentication​

Cubeo pre-selects the simplest authentication that will work. You'll see one of these:

  • No authentication — public servers. Nothing else to fill in.
  • Bearer token — the server expects a single static token. Paste it once.
  • OAuth 2.1 (auto-register) — Cubeo can register itself with the server on the fly. Click Connect and a new tab opens for you to authorize.
  • OAuth 2.1 (shared vendor app) — an admin has already curated this integration; you can use it as-is.
  • OAuth 2.1 (your own OAuth app) — the provider requires per-team OAuth credentials. Paste the client ID and secret you created in the provider.

Click Connect when you're ready. For OAuth flows you finish the consent in the provider's tab and Cubeo brings you back automatically when done.

Step 3 — Review the imported tools​

Once connected, Cubeo:

  1. Reads the catalog of tools the server exposes.
  2. Turns each one into a Cubeo tool you can edit and customise.
  3. If you connected from an AI Agent's Tools tab, every tool is added to that agent right away. If you connected from Team → Integrations, the tools are ready to add — open any AI Agent's Tools tab to enable them.

You'll see a success toast like "Imported 5 tools" and the page refreshes.

Each tool appears in the AI Agent's Tools list with an MCP badge, and any hints the server advertises (e.g. Read-only, Destructive).

Reusing the same MCP across multiple agents​

The connection is team-wide. To add the same MCP server's tools to another AI Agent, just open that agent's Tools tab and pick them from the catalog — you don't go through the Connect MCP wizard again. Each agent keeps its own copy of the tools, so renames, custom descriptions, and parameter overrides on Agent A don't affect Agent B.

Step 4 — Customise (optional)​

Click an imported tool's Configure button to:

  • Rename the tool so the agent has a clearer mental model (e.g. search_repos becomes Find a GitHub repo).
  • Edit the description the agent reads when deciding whether to call it.
  • Hide or expose individual parameters from the agent.
  • Tweak parameter help text or default values.

Your changes only affect this AI Agent — the upstream MCP tool definition is untouched.

Managing the connection later​

Sync, edit, remove, or troubleshoot an MCP server from Team → Integrations. See Manage your connections for the lifecycle operations and the MCP reference for the troubleshooting matrix.