🔧 MCP Tool

Overview

The MCP Tool node lets you call tools from connected Model Context Protocol (MCP) servers. MCP provides a standardized way to discover and invoke tools across different AI platforms and services.

Configuration

• MCP Server — select a configured MCP server connection
• Tool Name — the specific tool to invoke from the server
• Tool Parameters — input parameters as JSON (supports {{variables}})
• Output Variable — name for storing the tool response

Supported Transports

• Streamable HTTP — modern MCP HTTP transport with optional SSE upgrades. The recommended choice for new servers.
• SSE — older MCP server-sent events transport. Still common.
• HTTP Legacy — Falcon's pre-MCP-spec REST shape (POST /tools/execute). Existing connections continue to work unchanged; new connections should pick Streamable HTTP or SSE.

Authentication

• None — no auth headers
• Bearer token — sends Authorization: Bearer <token> from a workspace credential
• Custom header — sends a configurable header name + value from a workspace credential

Use with Agent Loop

MCP servers connected to the workspace can also be referenced by Agent Loop nodes via the "MCP Servers" multi-select on the agent-loop config panel. The agent will see every cached tool the server exposes and can call them autonomously across multiple turns.

Features

• Tool Discovery — "Discover Tools" refreshes the cached tool catalog. The cache is reused on every workflow run to avoid an extra round-trip.
• Secure Invocation — tools are called with proper authentication and sandboxing
• Governance & Audit — all tool invocations are logged for compliance and debugging