Skip to main content
Version: V11

MCP Node

The MCP Node connects workflows to external Model Context Protocol (MCP) servers, enabling access to tools and services hosted on remote systems. MCP is an open standard that provides a unified way for AI systems to connect with external data sources, APIs, and specialized functionality.

The node operates in two modes: as a standalone node with user-provided arguments, or as a tool connected to an AI node where the model provides arguments at runtime.

How It Works​

The MCP Node establishes a connection to an MCP-compatible server using the configured transport protocol (Streamable HTTP, SSE, or WebSocket). Once connected, the node can discover and invoke tools exposed by the server. The server processes tool requests and returns results to the workflow.

When used as a standalone node, the MCP Node executes tools with arguments specified in the configuration. When connected to an AI node through a green connector, the AI model decides which tools to invoke and provides arguments based on user input. This dual-purpose design supports both deterministic workflows and dynamic AI-driven interactions.

The node caches tool discovery results to improve performance. When server configuration changes (URL, transport, or headers), the cache invalidates automatically to ensure fresh tool definitions.

Configuration Parameters​

Server Name​

Server Name (Text, Required): A unique identifier for the MCP server connection.

This name prefixes tool names when multiple MCP servers are connected to the same workflow. The prefix helps the AI distinguish between tools from different sources. Choose a descriptive name that reflects the server's purpose.

Examples: weather-service, order-management, document-search

Server URL​

Server URL (Text, Required): The endpoint URL of the MCP server.

Enter the complete URL including the protocol. The URL format depends on the selected transport protocol. Variable interpolation with ${variable_name} is supported for dynamic endpoint selection.

Examples:

  • https://mcp.example.com/
  • https://api.service.com/mcp
  • ${config.mcp_endpoint}

Transport​

Transport (Dropdown, Required): The communication protocol for connecting to the MCP server.

ProtocolDescription
Streamable HTTPStandard HTTP-based transport for request-response communication. Recommended for most integrations.
SSEServer-Sent Events transport. Legacy protocol for older MCP servers that do not support Streamable HTTP.
WebSocketBidirectional persistent connection. Suitable for high-frequency tool invocations or servers requiring WebSocket connectivity.

Headers​

Headers (Key-Value Pairs, Optional): HTTP headers sent with each request to the MCP server.

Headers provide authentication credentials and configuration values required by the server. Each header requires a name and value pair.

Common patterns:

  • Authorization: Bearer <token> — OAuth or JWT authentication
  • X-API-Key: <key> — API key authentication
  • Content-Type: application/json — Request format specification

Store sensitive credentials in workflow variables rather than entering them directly. Reference variables using ${variable_name} syntax.

Tools​

Tools (List, Optional): Specific tools from the MCP server to expose.

When no tools are specified, all tools exposed by the MCP server become available. When specific tools are listed, only those tools can be invoked.

Each tool configuration includes:

  • Name: The tool name as defined by the MCP server
  • Arguments: Parameters for the tool (node mode only). Variable interpolation with ${variable_name} is supported.

Tool filtering limits AI access to approved operations, reduces complexity by exposing only relevant functionality, and enforces security boundaries around sensitive operations.

Output Field​

Output Field (Text, Required): Workflow variable where tool results are stored.

Results are accessible to other nodes using ${state.data.<variable_name>} syntax. The output format depends on the MCP server's tool implementation.

Default: mcp_result

Execution Lifecycle​

The MCP Node executes through these stages:

  1. Connection: The node establishes a connection to the MCP server using the configured transport protocol
  2. Tool Discovery: Available tools are fetched from the server (cached for performance)
  3. Tool Selection: In tool mode, the AI selects which tool to invoke. In node mode, configured tools execute directly.
  4. Argument Resolution: Arguments are resolved from configuration (node mode) or AI input (tool mode). Variable interpolation is applied.
  5. Tool Execution: The selected tool executes on the MCP server
  6. Result Storage: Server response is stored in the specified output field

MCP Node vs Tool​

The MCP Node operates in two distinct modes based on how it connects to the workflow:

Node Mode (Orange Connector)

  • Executes as part of sequential workflow flow
  • Arguments are provided in the node configuration
  • Tools execute deterministically based on configuration
  • Suitable for predictable, repeatable operations

Tool Mode (Green Connector)

  • Connected to an AI node through the green connector
  • AI model decides when to invoke tools
  • AI provides arguments based on user input
  • Suitable for dynamic, context-aware operations

Common Parameters​

This node supports common parameters shared across workflow nodes, including Stream Output Response, Logging Mode, and Wait For All Edges. For detailed information, see Common Parameters.

Best Practices​

  • Use descriptive server names to identify tool sources in workflows with multiple MCP connections
  • Limit tool exposure to only the operations needed for the workflow
  • Store authentication credentials in secure workflow variables
  • Test MCP server connectivity before deploying workflows to production
  • Configure appropriate timeouts for external service calls
  • Use Streamable HTTP transport unless the server specifically requires SSE or WebSocket

Limitations​

  • Server Compatibility: The target MCP server must implement the Model Context Protocol specification and support the selected transport protocol.
  • Network Dependency: Tool execution requires network connectivity to the MCP server. Network latency affects response times.
  • Authentication Requirements: Server-specific authentication must be configured correctly. Invalid credentials result in connection failures.
  • Tool Availability: Available tools depend on what the MCP server exposes. Tool changes on the server may require workflow updates.
Did you find this article helpful?