ADD MCP (Model Context Protocol) Server to Support AI Query

ADD MCP (Model Context Protocol) Server to Support AI Query

Release Date: 08/15/2025
Version: 8.6 & up

One Elliott user had asked if we plan to create a MCP (Model Context Protocol) Server.  Currently, we do have the plan.  However, if we do so, then potentially we can support the following AI question:
  1. Check stock for item code ABC123
  2. Place order for 10 units of widget W1 to customer 100045
  3. What’s the balance for customer Smith Hardware?
  4. Send invoice for order 22589
  5. how many of X did we sell yesterday
  6. who is (or was yesterday, last month, quarter, year…) our largest dealer in terms of orders and/or dollars
  7. what are they buying the most
  8. when do you think they’ll place their next order
  9. what was our contribution margin yesterday
  10. what dealer/AE gave us the largest/least margin…
  11. what’s my avail credit
  12. raise/lower credit limit
  13. pay invoice x
  14. show me my AEs by sales (num orders, dollars)
  15. show me dealers who have not purchased in the past X days, weeks, quarter, year
  16. compare dealer x year over year…
Per ChatGPT, in addition to build the MCP server, we need the following:

Great question — you’re really describing the "last mile" between humans asking business questions in natural language and your MCP server (which sounds like it would be the backend intelligence layer hooked to your ERP data).

To make this work, you’ll need a front end that can:

  1. Accept natural language input (typed or spoken).

  2. Translate it into structured queries/commands that the MCP server understands.

  3. Display the results back in a useful way (text, dashboards, tables, charts, alerts, etc.).

Here’s a breakdown of the main front-end approaches you could consider:

🔹 1. Conversational UI (Chatbot / Virtual Assistant)

  • UI Layer: A chat window (web, desktop, or mobile app) where a user types “Who was our largest dealer last quarter?”

  • Middleware: A natural language processing (NLP/LLM) component that maps the free-form text into MCP server API calls.

  • Result: MCP server returns structured data → chatbot formats into plain language, tables, or charts.

👉 Example: Think of Slack/Teams bot, or a web-based “Ask ERP” assistant.

🔹 2. Voice UI

  • Same as above, but adds speech-to-text in front and possibly text-to-speech on the way out.

  • Useful for execs who want to “ask ERP” from their phone or car.

🔹 3. BI/Analytics Dashboard with Natural Language Queries

  • Embed NLP into a dashboard (similar to Power BI’s “Q&A” or Tableau Ask Data).

  • Users can type or select guided queries, which are parsed and sent to MCP.

  • Results can come back as interactive charts/tables.

  • This works well for recurring queries like “sales by dealer YTD vs last year”.

🔹 4. Hybrid UI (Chat + Widgets)

  • Chat for unstructured questions.

  • When the MCP server detects certain query types (like “pay invoice X” or “raise credit limit”), it can return action widgets (buttons/forms).

  • Example:

    • User: “Pay invoice 12345.”

    • MCP: “Invoice 12345: $12,500 due. Confirm payment?” [Yes/No button].

This keeps things user-friendly and prevents accidental destructive commands.

🔹 5. ERP Native Integration

  • If your ERP already has a web/desktop UI, you could embed an “Ask MCP” bar directly inside.

  • That bar sends natural language to MCP → returns structured output.

  • The ERP UI already knows how to show invoices, dealers, credit limits, so MCP can just trigger those views with the right parameters.

🔧 Tech Pieces Needed:

  1. NLP/LLM Layer (OpenAI, Azure AI, AWS Bedrock, Rasa, etc.) → maps “business English” into MCP query schema.

  2. Orchestration Layer (your “translator”) → turns LLM output into MCP API calls.

  3. UI Layer → Chatbot, dashboard, voice, or ERP-embedded input.

  4. Visualization Components → Tables, charts, notifications, or action buttons.

✅ Summary:
The most natural front end for your MCP server is a conversational UI (chat or voice) backed by NLP, with BI/dashboard components for structured visualization and action widgets for commands like paying invoices or adjusting credit. This hybrid approach gives executives an “ask anything” experience while keeping operational tasks safe and controlled.

Here’s the sample architecture diagram showing how a user’s natural language question flows through the UI → NLP → Orchestration → MCP server → ERP data, and then back to the front end with results or actions.