MCP Servers

Model Context Protocol (MCP) enables connecting external context servers to extend AI model capabilities.

Overview

MCP servers provide additional tools and context for AI agents, allowing them to:

• Access specialized functions from external servers
• Integrate with systems via the MCP protocol
• Extend context available to the LLM

Viewing MCP Servers

To manage MCP servers:

1. Go to Integrations
2. Switch to the MCP Servers tab

Adding an MCP Server

1. Click Create MCP server
2. Fill in the form:
   • Name: Server name (required)
   • Description: Purpose of the server
   • URL: Server address (required)
   • Type: Connection type (STREAMABLE_HTTP or SSE)
   • Use OAuth: Enable if the server requires OAuth authorization
     • Client ID: Client identifier
     • Client Secret: Client secret key
   • Headers: Additional HTTP headers (key-value pairs) needed for connection
3. Click Create

Managing MCP Servers

From the server list you can:

• Edit settings for existing servers
• Delete servers that are no longer needed

Using MCP Tools in Agents

Once an MCP server is configured:

1. Open your AI agent in Agentflow
2. Add an MCP tool node to the canvas
3. Connect it to the AI Agent node's tool handler (purple port)
4. The agent can now use tools provided by the MCP server