JDE Metadata MCP Server
Your JD Edwards data model, served to any MCP-capable AI assistant as an instant, authoritative reference — data items, tables, keys, code values, views, functions and their parameter structures — without giving the assistant your database.
AI help with JDE is only as good as its knowledge of your data model
Ask a general-purpose assistant about JD Edwards and it will answer fluently — and guess. Column names, code values, parameter lists and key structures are exactly the things a language model invents most confidently, because they look plausible and cannot be derived from prose. One invented column name costs more time than the assistant saved.
The JDE Metadata MCP Server removes the guessing. It serves the actual data model of your system — extracted from it, once, on your side — so an assistant can look a fact up instead of inventing it: what a data item means and how it is stored, what a table contains and how it is keyed, what a code value stands for, what a business function takes and returns.
- Data items, tables, keys and indexes, code types with all their values, business views, business functions and their parameter structures.
- Reverse lookups: where an item is used, which views read a table, what a function module contains.
- Read-only, and no database connection at runtime.
- Metadata extraction from your own system is included.
- Immediate responses, served entirely from memory.
- STDIO and Streamable HTTP editions.
Every layer of the model, both directions
Look a fact up
A data item’s meaning, storage type and display rules. A table’s columns, prefix and description. Its keys and their column order. A code type with every value it holds. A view’s member tables, columns and joins. A function’s parameters — names, types, direction, required or optional.
Ask the reverse question
Which tables carry this data item? Which views read this table? Which functions live in this module, and which of them use this parameter structure? Reverse lookups answer these exactly, where text search would only approximate.
Search when the name is unknown
Find an identifier from a partial description, a fragment of a name, or a pattern — in one category of metadata or across all of them at once, for the “what is this thing even called” moments.
A reference, not a connection
The server holds a compact, encrypted copy of your system’s metadata and answers from memory. At runtime it needs no database connection, runs no SQL, and holds no JDE credentials — there is nothing in the AI’s path that could read business data, because the ability simply is not there. Responses are immediate for the same reason.
The extraction utility is included: point it at your own JD Edwards system once and it packages that system’s metadata for the server. Your metadata is extracted on your side and stays on your side.
Guidance built in
Raw reference data is necessary but not sufficient — an assistant also has to interpret it the JDE way. Dates that are not stored as dates, numbers whose decimal places are implied rather than stored, values that are padded before they are compared: JD Edwards has conventions that generic knowledge gets wrong.
So the server ships with several levels of helpful guiding for the target LLM, delivered alongside the answers where it is relevant. The result is an assistant that not only sees the correct definition but reads it correctly — and produces work that respects how the data is actually stored.
Two editions, one server
The STDIO edition plugs straight into desktop MCP clients — the pattern used by Claude Desktop, IDE assistants and local agent runners. The Streamable HTTP edition serves the same tools over HTTP or HTTPS for clients that connect over the network, and is designed to sit beside your other services on existing endpoints.
Both editions serve the same metadata with the same guidance. For remote enterprise AI clients such as Microsoft Copilot, Steward can front the server and keep access bound to the identity of the person asking.
The sort of thing an assistant stops guessing
- “What does this column actually store, and how many implied decimals does it have?”
- “Which index should this query use, and in what column order?”
- “List the valid values for this code and what each one means.”
- “What parameters does this business function take, which are required, and what structure do they travel in?”
- “Where else is this data item used — which tables, views and structures would a change touch?”
- “I only remember part of the name — find the table that holds supplier invoices.”
Background reading
Want your assistants grounded in your data model?
Tell us which assistants your teams use and where they lose time to guessed names and misread values. We can show the reference working against real metadata.