If you are an Appian developer who spends a lot of time inside an AI-powered editor or chat assistant, you know how tedious it is to switch back and forth between your IDE and the Appian Designer to look up an object, verify a record type, or run an expression rule.

Appian MCP Server changes that. It lets system administrators, developers, and end users connect external AI applications to Appian capabilities through Model Context Protocol (MCP). Once enabled, MCP-compatible AI applications can talk to your Appian Cloud environment using a service account API key and interact with application data and business logic through natural language.

What is Model Context Protocol?

Model Context Protocol (MCP) is an open industry standard for connecting AI applications to external tools and data sources. It 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. Appian supports it the same way it supports REST APIs or SAML for authentication — as an adopted standard that extends Appian’s reach into the broader AI ecosystem.

What the Appian MCP Server makes available

Through MCP, an AI application can access the following Appian capabilities:

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

Generic tool architecture

Rather than creating a separate MCP tool for every design object, the Appian MCP Server uses a small, fixed set of system tools that handle discovery and execution for all MCP-enabled 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.

The developer workflow

The developer experience is refreshingly simple:

  1. Open a design object (like an expression rule or process model) in Appian Designer.
  2. Enable the MCP tool checkbox in the object’s properties — no code changes or redesign required.
  3. Write a clear tool description so the AI application knows when to pick your tool.

That is it. Your design object is now discoverable and invokable from any connected AI application.

Security model

Connections authenticate as a single service account using an API key. The service account’s permissions determine what is accessible:

  • Object security: the service account can only discover and invoke tools for objects it has viewer access to.
  • Record-level and field-level security: data fabric queries respect the security configured on the record type. Records and fields the service account cannot access are never returned.
  • 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, the tool invoked, the timestamp, and the outcome.

Limitations to keep in mind

  • Appian Cloud only — self-managed environments are not supported.
  • No document upload — external AI applications can retrieve documents but cannot upload them.
  • No start forms — 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 — there is no per-user authentication.
  • Synced record types only — record types using direct data access or a record-level security expression are not queryable.

Getting started

Each role has its own entry point:

Now your AI assistant can query record data, run expression rules, kick off process models, and even invoke AI agents — right from the tool you already use. Have you tried connecting an AI tool to your Appian environment yet?