MCP Integration

Model Context Protocol (MCP) lets you connect external tool servers and expose their capabilities to your workflows and agents.

What is MCP?

MCP is an open protocol that standardizes how applications provide context to LLMs. In this platform, MCP servers are managed through Settings. Each server can use stdio or streamable-http transport. Once connected, MCP tools are listed automatically and can be invoked from the MCP workflow step or the MCP API.

Configuration

MCP is configured in Settings under Developer Integrations. You can add multiple servers, enable or disable them individually, and choose between stdio and streamable-http transports. The backend also supports environment-based configuration via .env.

# Enable MCP globally
MCP_ENABLED=true

# Optional: path to a JSON config file
MCP_CONFIG_PATH=/path/to/mcp-config.json

# Optional: inline JSON config
MCP_CONFIG_JSON={"servers":[...]}

# Optional: default server URL for streamable-http
MCP_SERVER_URL=http://localhost:3001/mcp

Server Configuration

Each MCP server requires a unique ID, a name, a transport, and transport-specific fields. The autoDiscover flag controls whether the platform automatically lists the server's tools.

{
  "id": "my-mcp-server",
  "name": "My MCP Server",
  "transport": "stdio",
  "command": "node",
  "args": ["/path/to/server.js"],
  "url": "",
  "headers": {},
  "env": {},
  "enabled": true,
  "autoDiscover": true,
  "timeoutMs": 30000
}

MCP Workflow Step

Use the mcp step type in a workflow to call a tool exposed by an MCP server. The step requires the server ID, tool name, and arguments.

{
  "type": "mcp",
  "serverId": "my-mcp-server",
  "toolName": "search_docs",
  "arguments": { "query": "{{last}}" },
  "timeoutMs": 30000,
  "maxRetries": 2,
  "backoffMultiplier": 2
}

MCP API Reference