MCP Node

What it does

The MCP node connects your AI product to any server that speaks the Model Context Protocol. Once connected, Claude can call that server's tools at any point in a conversation — looking up live data, writing back to your CRM, checking availability, or running any action the server exposes.

This is a two-way, conversational integration. Unlike a webhook (which fires once when a conversation ends), MCP tools are available to Claude throughout the session. The AI decides when to call them based on context.

How to add it

1. Drag the MCP node from the canvas toolbar onto your canvas
2. Click it to open the settings panel
3. Enter your Server URL — the MCP endpoint (e.g. https://mcp.yourserver.com/mcp)
4. Optionally enter a Bearer Token if your server requires authentication
5. Click Test Connection — Productised fetches the server's tool list and shows every available tool
6. Enable or disable individual tools using the checkboxes — limit Claude to only the tools relevant to this product
7. Give the node a Display Name so you can tell it apart if you add multiple MCP nodes

What Claude can do with MCP tools

Read from your systems mid-conversation:

• Look up a prospect's existing CRM record before asking questions they've already answered
• Check live pricing or availability before making a recommendation
• Pull in account history to personalise the conversation

Write back to your systems:

• Create a contact record when the user provides their details
• Log a conversation note or summary to the CRM
• Update a deal stage based on what the AI collected

Post-conversation actions via system prompt: You can instruct Claude in the system prompt to call specific tools when collection is complete:

"When you have collected the user's name, email, and company, call create_contact with those details before giving your recommendation."

This makes MCP tools behave like a guaranteed post-completion action — similar to a webhook, but driven by the AI rather than the platform.

What makes MCP different from the webhook

Both can coexist. A product can have an MCP node for live CRM lookups during the conversation and a webhook for guaranteed post-completion data routing.

Supported MCP servers

Any server implementing the MCP Streamable HTTP transport (JSON-RPC 2.0 over HTTP POST). This includes:

• Servers built with the official @modelcontextprotocol/sdk
• Hosted MCP servers from tools like Affinity, Notion, Linear, and others
• Your own custom server built with any language that can serve HTTP

The server must expose a tools/list endpoint (used by Test Connection) and a tools/call endpoint (used at runtime). Bearer token authentication is supported.

Per-product isolation

Each product on your canvas has its own MCP node with its own server URL, token, and enabled tool list. A coaching product can connect to your coaching methodology server, a sales product to your CRM — completely independent.

Coming: multiple MCP nodes

A single product can have multiple MCP nodes connected — one for your CRM, one for your calendar, one for your knowledge base. Each contributes its tools to Claude's available tool set for that conversation.