Set-up MCP Connector

Connect your Wonderchat chatbot to a remote MCP (Model Context Protocol) server so it can call that server's tools mid-conversation.

The MCP Connector lets you connect a Wonderchat chatbot to a remote MCP (Model Context Protocol) server, so the bot can call that server's tools in the middle of a conversation.

Like custom Tools, this gives your chatbot live, real-world capabilities beyond its knowledge base — for example, looking up or updating records in a CRM, ticketing system, or internal tool via that system's MCP server. The available tools are whatever the connected server exposes; nothing is hard-coded on the Wonderchat side.

Wonderchat acts as an MCP client — it connects out to an external MCP server. It does not expose your chatbot as an MCP server for others to connect to.

Plan availability

The MCP Connector is available on the Scale plan and above. On lower plans the "Connect MCP Server" card is visible but dimmed, with a plan badge. If a workspace downgrades below Scale, existing MCP connections are automatically disabled.

How to connect an MCP server

Navigating to the MCP Connector

1. From the chatbots page, open Edit Chatbot for the bot you want to configure.

2. Within the edit chatbot menu, click on the "Tools" tab.

3. On the Tools page, find the "Connect MCP Server" card.

Setting up the connection

Setting up a connection is a short, step-by-step wizard:

Step 1: Name the connection

• Connection Name: Provide a name to identify this connection (e.g. HubSpot).

• Remote MCP Server URL: Paste the URL of the remote MCP server (e.g. https://mcp.hubspot.com).

The server URL must be a public HTTPS address. localhost, .local / .internal hostnames, and private IP addresses are blocked for security reasons.

Step 2: Authenticate

Choose how Wonderchat should authenticate with the server. Two modes are supported:

• OAuth 2.0 — Wonderchat runs the OAuth flow (with PKCE) against the server, including discovery and dynamic client registration where the server supports it. You'll be prompted to authorize the connection. Tokens are refreshed automatically if a call comes back unauthorized.

• Custom header (API key) — Provide a static header the server expects, for example Authorization: Bearer <key> or X-API-Key: <key>.

Your secrets are never sent back to the browser. The interface only ever shows connection status (connected / not connected) — never the stored credentials themselves.

Step 3: Pick the tools

Once connected, Wonderchat lists the tools the server exposes. Choose which of those tools you want to make available to your chatbot.

Configuring each tool

Each tool you enable has its own settings:

• Require confirmation (on by default): When on, the chatbot asks the user to confirm before the tool actually runs.

• When-to-use guidance: Guidance fed to the model that tells it when this tool should be used.

• Tool-called status message: A custom message shown to the user when this tool is called.

Managing connections

Existing connections appear in a list, each with an on/off toggle so you can enable or disable a whole connection without deleting it.

How it works at chat time

• Transport: MCP connections use Streamable HTTP only (SSE and stdio transports are not supported).

• During a conversation: Each enabled tool is made available to the chatbot's agentic loop. The AI model decides when to call a tool based on the conversation and your when-to-use guidance; the tool's text result is returned to the model to continue its answer.

• Logging: Every tool call is logged — as an action in the chatlog, and as a tool run for analytics — so you can review what was called and when.

• Short-lived: Connections are opened per call and are not held open between messages.

---

Related

Set-up Custom Chatbot Tools