The following steps outline how a user request is processed from initiation to response within the OpenEdge MCP Server architecture:
  1. The end-user interacts with an application, for example an IDE or a chat interface through an AI Agent, which delegates the user query to the MCP Client.

    For example, the end-user enters a natural language request, such as: “Show me all the available inventory.”

  2. The MCP Client authenticates to the MCP Server using both service account and user credentials. When operating in OAuth2 mode, the MCP Server functions as an authorization server. It delegates user authentication to external Identity Providers using OAuth2 flows, including the authorization code process with PKCE for enhanced security. During initialization, both parties declare their supported capabilities:
    • Tools – Functions the server can execute
    • Resources – Data the server can provide
    • Prompts – Templates offered by the server that guide AI responses and can coordinate multiple tools for complex workflows
    The MCP Server lists tools and prompts allowed by the scope of the user. Once initialized, the connection remains persistent. Both, MCP Client and Server are responsible for handling errors and maintaining the connection state.
  3. The MCP Client forwards the query and server information to the LLM, which analyzes the request and recommends the appropriate MCP tools for execution.
  4. The MCP Client executes the recommended tools with necessary parameters on the MCP Server.
  5. On invocation, the MCP Server optionally exchanges tokens with PAS for OpenEdge SSO.
  6. The MCP Server calls PAS for OpenEdge Services with sanitized headers.
  7. Response guard enforces limits before returning results to the MCP Client.
  8. The MCP Client returns the tool results to the AI Agent, and the LLM incorporates the new information into its reasoning. The LLM may decide to call additional MCP tools if needed, repeating steps 4–7 as necessary.
  9. The LLM generates the final response using all available contexts.
  10. The AI Agent presents the LLM's final response to the end-user.

    For example, as requested by the end-user in step 1, the inventory summary may appear as:

    Inventory Summary:

Raw Materials: 3,890 units
Work in Progress: 246 units  
Finished Goods: 585 units

Low Stock Alerts: Aluminum Rods, Model X-300

Total Value: $2.4M
Last Updated: Sep 15, 14:45