MCP Tools for AI Agents

This page explains how to connect AI agents to external tools provided by Model Context Protocol (MCP) servers. You'll learn how to set up an MCP connected system, add MCP tools to an AI agent, and manage tool changes over time.

What are MCP tools?

Model Context Protocol (MCP) tools are external tools provided by an MCP server that AI agents can call at runtime. Unlike native Appian tools—such as record types, process models, or other AI agents—MCP tools run outside of Appian on a third-party server.

MCP tools let your AI agents interact with external systems like Snowflake, GitHub, Google Drive, and Box without building custom integrations. You connect to the MCP server once using an MCP connected system, then select which tools to make available in each AI agent.

Set up an MCP connected system

Before you can add MCP tools to an AI agent, you need an MCP connected system that points to an active MCP server. See MCP Connected System for setup instructions.

Add MCP tools to an AI agent

Adding MCP tools to an AI agent uses a two-step selection flow:

1. In Appian Designer, go to the Tools section of your AI agent.
2. Click Add. Result: The Add Tools screen displays.
3. Select the EXTERNAL TOOLS tab.
   

   Scroll or drag to pan · ⌘+/⌘− to zoom · Esc to close

4. Select the MCP connected system for the server you want to use or click ADD SERVER to create a new connected system.
5. Choose your tools. You can select all tools from the server or pick specific ones.
   

   Scroll or drag to pan · ⌘+/⌘− to zoom · Esc to close

6. Click ADD.

The selected MCP tools appear in the tools list alongside any native Appian tools.

Discover available MCP tools

When you select an MCP connected system, your AI agent connects to the server and displays the tools the server offers. Each MCP tool displays its name and a description of what it does. Tool names and descriptions come directly from the MCP server.

If the MCP server is unavailable when you try to add tools, an error displays. Check the server URL and credentials in the MCP connected system configuration.

MCP tool execution at runtime

When an AI agent runs, the AI agent can call MCP tools the same way it calls native Appian tools. The AI agent retrieves the tool definitions—including names, descriptions, and input schemas—from the MCP server, decides which tool to use, and sends a request. The MCP server runs the tool and returns results to the AI agent.

If an MCP tool call fails at runtime, the AI agent receives an error message from the MCP server. The AI agent can then decide how to proceed—retry, try a different tool, or stop.

MCP tool calls do not consume additional AI actions. The cost is the same as any other tool type.

Detect MCP tool changes

MCP servers can change or remove tools at any time. Appian detects MCP tool changes in the following ways:

• Design time: If an MCP tool becomes unavailable, an error icon appears on the tool card when the AI agent configuration reloads.
• Runtime: If the AI agent tries to use an MCP tool that no longer exists, the AI agent receives an error during the tool call step because Appian cannot retrieve the tool definition.

MCP tool security and access control

Access to MCP tools is governed at two levels:

• Connected system security: Only developers with Viewer permissions or higher on the MCP connected system can select the MCP server in an AI agent. Connected system security controls which developers can use the MCP server.
• Agent-level configuration: The AI agent configuration controls which specific MCP tools from the server are available during execution. Even if a server offers 20 tools, you can limit the AI agent to only the MCP tools it needs.

MCP tool limitations

MCP tools have the following limitations:

• Tools primitive only: MCP defines three primitives: tools, resources, and prompts. Appian supports the tools primitive only.
• Authentication methods: Only API key and OAuth 2.0 Client Credentials Grant are supported for MCP server connections.
• No automatic tool migration: If the MCP server changes or removes a tool, you must update the AI agent configuration manually.
• No server or tool count limits: Appian does not impose limits on the number of MCP servers or MCP tools per agent. However, the 15-tool recommendation applies to the total number of tools (native and MCP combined).
• Server-side rate limits: Appian does not limit MCP tool calls, but the external MCP server may enforce its own rate limits. Check the MCP server vendor documentation for details.

Feedback