Appian MCP Server

Appian MCP Server lets system administrators, developers, and end users connect external AI applications to Appian through Model Context Protocol (MCP). When enabled, users can connect MCP-compatible AI applications to their Appian Cloud environment to query application data, start processes, evaluate expression rules, and call AI agents through natural language.

This page explains the Appian MCP Server, its benefits, and how your data is secured.

What is Model Context Protocol?

Model Context Protocol (MCP) is an open industry standard for connecting AI applications to external tools and data sources. MCP defines how an AI application connects to an MCP server to discover available tools, understand what inputs those tools expect, and invoke them.

MCP is not Appian-specific technology. Appian supports MCP the same way it supports REST APIs or SAML for authentication—as an adopted standard that extends Appian's reach into the broader ecosystem.

Why Appian supports MCP

Model Context Protocol (MCP) provides a standards-based way for AI applications to discover and use Appian data and business logic without custom API development for each integration. Customers use third-party AI applications to interact with business data and automate tasks through natural language.

MCP also acts as a two-way interface. External AI applications can access Appian data and business logic through Appian MCP Server, and Appian AI agents can use MCP to access external tools and data sources outside the platform. Appian also supports MCP connected systems, which allow Appian AI agents to leverage external MCP servers as tools within the platform.

Benefits of Appian MCP Server

Appian MCP Server provides the following benefits over building custom integrations:

• Standards-based integration: A single MCP server provides a standards-based integration point for any MCP-compatible AI application, eliminating the need for custom API development.
• Automatic security enforcement: Appian's existing security model applies automatically to all MCP connections without building a separate security layer.
• Generic tool architecture: A small, fixed set of system tools can discover and call any MCP-enabled design object, scaling to hundreds of tools without overwhelming the AI application's context window.
• Zero-rebuild enablement: Developers enable existing design objects as MCP tools by selecting a checkbox in properties. No code changes or redesign required.
• Portable configurations: MCP tool configurations deploy with design objects through packages, eliminating manual re-enablement across environments.

How MCP fits with Appian AI capabilities

MCP is complementary to other Appian AI capabilities, not a replacement. Use the following guidance to determine which capability fits your use case:

Capability	When to use
Appian AI agents	You want an AI agent embedded in your Appian application that end users interact with through Appian's interface.
Composer	You want AI assistance for planning, building, and modifying design objects in Appian Designer.
Dev MCP	Developers want to use a third-party IDE with AI to read and write Appian design objects. Dev MCP is a community-supported App Market plug-in, not part of Appian MCP Server.
Appian MCP Server	Users already work in a third-party AI application and need to query Appian data, start processes, or call AI agents from that application.

All of these capabilities enforce Appian's existing security model, including object security, record-level security, and group membership. For details on how Appian MCP Server applies these controls, see Appian MCP Server Security.

What Appian MCP Server makes available

Appian MCP Server provides access to the following Appian capabilities through MCP:

• Record types: Query data from synced record types using SQL.
• Process models: Start process models that are available as MCP tools and check process execution status.
• Expression rules: Evaluate expression rules that are available as MCP tools with provided inputs.
• AI agents: Invoke Appian AI agents that are available as MCP tools and check agent execution status.

Supported AI models

Because Appian MCP Server connects your external AI application to Appian, you can use any AI model your AI application supports. Appian does not restrict, validate, or select the model for MCP connections. The model is determined entirely by the AI application you connect, so Appian MCP Server works with any current or future model that application offers.

Model	Provider

Generic tool architecture

Appian MCP Server uses a generic tool architecture rather than creating a separate MCP tool for each design object. A small, fixed set of system tools handles discovery and execution for all MCP-enabled design objects:

1. Search: The AI application searches for available tools matching the user's request.
2. Get schema: The AI application retrieves the full input and output schema for a specific tool.
3. Invoke: The AI application calls the appropriate system tool to execute a process model, expression rule, or AI agent.
4. Query: The AI application analyzes the available record types and securely queries the requested data.

This architecture scales to hundreds of developer-enabled tools without overwhelming the AI application's context window. For the complete tool reference, see MCP System Tools Reference.

Security model

Appian MCP Server uses service account API key authentication for all MCP connections. All connections authenticate as a single service account rather than individual end users. The service account's permissions determine what data and tools are accessible through MCP:

• Object security: The service account can only discover and invoke MCP tools for design objects it has Viewer permission to.
• Record-level and field-level security: All data fabric queries respect the record-level security and field-level security configured on the record type. Records and fields that the service account cannot access are not returned to external AI applications.
• Group membership: Tool visibility is filtered by the service account's group membership.
• Audit trail: Every MCP tool call is logged with the service account, tool invoked, timestamp, and outcome.

For the full security reference, see Appian MCP Server Security.

Role-based entry points

Appian MCP Server involves three roles, each with a distinct workflow for MCP access: System administrators, developers, and end users.

System administrators

System administrators enable Appian MCP Server in the Admin Console, configure authentication, and manage which tool collections are available.

To get started, see Enable and Configure Appian MCP Server.

Developers

Developers make new or existing design objects available as MCP tools and ensure those objects have effective descriptions that help AI applications select the right tool.

To get started, see Make Design Objects Available as MCP Tools.

End users

End users connect their AI application to Appian MCP Server using configuration details from their user settings. Once connected, they interact with Appian data and business logic through natural language.

To get started, see Connect Your AI Application to Appian MCP Server.

Appian MCP Server limitations

Appian MCP Server has the following MCP limitations:

• Appian Cloud only. Appian MCP Server is available only on Appian Cloud environments. Self-managed environments are not supported.
• No document upload. External AI applications can retrieve document content but cannot upload documents through MCP.
• No start form interaction. Process models with start forms cannot be invoked through MCP.
• No UI-dependent steps. Process models containing synchronous user input tasks cannot complete through MCP.
• Service account authentication only. Appian MCP Server authenticates all connections using a service account API key. There is no per-user authentication.
• Object permissions required. The service account must have appropriate permissions to each design object. For details, see Appian MCP Server Security.
• Synced record types only. Only synced record types can be queried through the data fabric tools. Record types that use direct data access or have a record-level security expression are not accessible.

Feedback