MCP Server Node
Connect to an external MCP server and invoke a tool. Use template variables like {{.variable}} in arguments to reference values from previous steps.
Overview
The MCP Server node connects to a remote Model Context Protocol server, fetches its available tools, and lets you invoke one within your workflow. This is how you integrate external services — Stripe, GitHub, databases, or any MCP-compatible server — directly into your flows without writing custom HTTP requests.
Configuration
Connection
| Field | Description | Required |
|---|---|---|
Transport Type |
How to connect to the server | Yes |
Server URL |
The MCP server endpoint | Yes |
Auth Type |
Authentication method | Yes |
Output Variable |
Variable name for the tool result | Yes |
Transport Types
| Type | Description |
|---|---|
SSE |
Server-Sent Events — persistent connection with real-time streaming |
Streamable HTTP |
Standard HTTP with streaming support |
Auth Types
| Type | Fields | Description |
|---|---|---|
None |
— | No authentication required |
API Key |
Header Name, API Key | Sends the key in a custom header (e.g. X-API-Key) |
Bearer Token |
Bearer Token | Sends a Bearer token in the Authorization header |
Tool Configuration
After connecting, the node fetches all available tools from the server. You then configure:
| Field | Description | Required |
|---|---|---|
Select a tool |
Pick from the tools exposed by the server | Yes |
Timeout (seconds) |
Max time to wait for the tool response | No (default: 30) |
Tool Arguments |
Dynamic fields based on the selected tool's schema | Varies |
Tool arguments are generated automatically from the server's tool definitions. Each argument shows its name, description, type, and whether it's required.
How It Works
- Configure the connection (transport, URL, auth).
- Click Connect & Fetch Tools to discover available tools.
- Select a tool from the dropdown.
- Fill in the tool arguments — use
{{.variable}}syntax to inject values from previous nodes. - The result is stored in the output variable for downstream nodes.
Example: Search Stripe Documentation
Flow Start → MCP Server (Stripe)
Tool: search_stripe_documentation
Args: question = "How to create a subscription"
→ Model (summarize result)
→ Simple Output
Example: Create a Product via MCP
HTTP Request (get product data) → MCP Server (Stripe)
Tool: create_product
Args: name = {{.httpRequest.name}}
description = {{.httpRequest.description}}
→ Simple Output
Example: Chain Multiple MCP Calls
MCP Server (fetch customers) → For Loop (each customer)
→ MCP Server (create invoice)
→ Delay (1 second)
Best Practices
- Use SSE transport when the server supports it — it's more efficient for long-running tool calls
- Store sensitive credentials (API keys, tokens) in the Vault and reference them with template variables
- Set appropriate timeouts based on the tool's expected response time
- Use the output variable name to clearly identify what data the tool returns (e.g.
stripeCustomers,githubIssues) - Test tools individually with the Test Tool button before wiring them into complex flows