How to configure MCP Tool for the Start node?

The MCP Tool trigger exposes the section as a tool for AI assistants (Cursor, Fastline, etc.) connected to the platform via MCP (Model Context Protocol). The AI can call the section, pass parameters, and receive the result.

When you need it

• The AI assistant needs to get data from ConnectiveOne (exchange rates, catalog, checks).
• You need to call already configured scenario logic from the AI client without duplicating code.
• You use Cursor or another application with MCP support.

What to know

• Input Schema — description of input parameters (names, types, required). The AI passes these when calling.
• Return Schema — description of the result and list of keys the platform returns from the section in the MCP response.
• Result constant — for the AI to receive useful data, the section must write the result to a special constant at the end (via "save to constant" action).
• Parameters in scenario — Input Schema parameters are automatically created as constants with prefix mcp_input_ and are available as {{ mcp_param_name }}.

Before you start

• You are signed in with integrator or administrator rights.
• You have a section with logic to call from the AI.
• The AI client supports MCP (Cursor, Fastline, etc.).

Steps

1. Select MCP Tool trigger type

1. Select the Start node on the canvas.
2. In Node Inspector, in the Trigger type field, select MCP Tool.

2. Fill in Input Schema

Input Schema defines the parameters the tool accepts from the AI.

1. In the Input Schema block, click Add parameter.
2. For each parameter specify:
   • Name — in Latin, e.g. currency, date, product_id.
   • Type — string, number, boolean, etc.
   • Required — whether the parameter is required.
   • Default value — optional.
   • Description — short description for the AI (recommended).

Example parameters:

Parameters appear as constants in the scenario (e.g. mcp_param_currency) and can be used in nodes as {{ mcp_param_currency }}.

3. Fill in Return Schema

Return Schema defines which result keys are returned from the section.

1. In the Return Schema block, specify Keys — list of key names in the response (e.g. rate, date, source).
2. Add Description — short description of the tool for the AI.

4. Write result to constant

Critical: Without writing the result, the MCP response will be empty.

1. At the end of the section logic, add a save to constant action or equivalent.
2. Write the result to a constant whose keys match the Return Schema.
3. Ensure Return Schema keys match the written data.

5. Copy configuration for AI client

Node Inspector has two buttons:

• Copy config for .cursor/mcp.json — configuration with authorization via headers (recommended).
• Copy config with URL token — URL with token in the query (if the client does not support headers).

Copy the appropriate option and add it to the MCP configuration in your AI client.

6. Connect the AI client

Detailed connection instructions: Connect AI assistant via MCP.

Briefly:

1. Open MCP settings in the AI client.
2. Add MCP server with URL and headers (or URL with token) from step 5.
3. Save and reload MCP.

What happens next

The AI assistant sees the section as a tool and can call it by name, passing parameters according to Input Schema. The response contains keys from Return Schema.

How to verify it works

• Check that the tool appears in the AI client.
• Call the tool with minimal parameters.
• If the response is empty — verify result writing to constant and Return Schema keys.

Common issues

Related materials

• What is the Start node
• Connect AI assistant via MCP
• Configure MCP Tool parameters
• Configure Start node triggers