AI MCP Servers
A growing collection of focused MCP servers for JD Edwards, controlled files and AI-assisted troubleshooting across large live logs.
The Model Context Protocol (MCP) is rapidly becoming the standard way for AI agents to call tools exposed by servers under your control. The win is structural: instead of one giant “do anything to my JDE” integration, the agent sees a small, named set of operations — each with a typed schema, each individually authorisable, each individually auditable.
The servers listed below are deliberately focused. Each has a clear job and a clear boundary. The collection now spans JD Edwards Orchestrations, controlled file workflows, high-speed investigation of large live logs across multiple servers and folders, and an instant, read-only reference for the JD Edwards data model itself.
- Typed, named tools per server — not a wildcard “agent shell”.
- Standard MCP transport — works with any MCP-compatible agent.
- JDE auth is server-side; agents never see credentials.
- Log analysis works across configured servers, folders and large files.
- JDE metadata is served as a read-only reference — no database access in the AI’s path.
- Use Steward when remote enterprise AI needs an identity-preserving gateway to existing MCP servers.
JDE AIS — Orchestration MCP Server
One MCP server, covering the three operations an AI agent actually needs to be useful against JDE Orchestrations:
- Discovery — the agent can list the orchestrations the configured JDE user is allowed to see, by name, with descriptions. No guessing, no hard-coded catalogue.
- Introspection — for any discovered orchestration, the agent can fetch its input schema (parameter names, types, required vs optional) and its output shape. This means the agent can build the call correctly the first time, instead of probing.
- Execution — the agent can run an orchestration with a constructed payload and receive the structured response, just as any other AIS client would. All standard AIS error semantics apply.
Login support. The server handles JDE authentication on the agent’s behalf: the agent never sees JDE credentials, and the server manages JDE sign-in and the session transparently. From the agent’s perspective, it just calls tools; from JDE’s perspective, the work happens under a real, traceable JDE user.
Why this matters. Orchestrations are already the right boundary between “arbitrary access to JDE” and “a curated, sanctioned set of operations”. Customers spend years building exactly that boundary. Exposing orchestrations via MCP turns that existing work into a clean AI-ready surface, without inventing a new one.
File I/O MCP Server
One MCP server for filesystem operations within a configured root directory. The agent can read, write, list and search files inside the allowed area, and nothing outside it. Standard MCP authorisation and auditing apply: the agent does not get an open shell, just a small named set of file-shaped tools.
Useful as a building block whenever an AI workflow needs to read inputs, produce outputs, or browse a working folder — alongside other MCP servers (JDE AIS above, or others) in the same agent session.
MCP Server for Log Analysis
Give an AI controlled search and reading access to large live logs spread across any number of configured servers and folders. The server is designed for iterative troubleshooting: search broadly for a clue, read the relevant surrounding lines, then follow the next clue without loading entire log files into the model.
It is particularly useful for application and integration problems where the evidence is distributed across several machines, logs are changing while the incident is being investigated, or centralising every diagnostic file would be unnecessary overhead.
JDE Metadata MCP Server
Serve the actual data model of your JD Edwards system — data items, tables, keys and indexes, code types with all their values, business views, business functions and their parameter structures — to any MCP-capable assistant as an instant, read-only reference, extracted from your own system with the included utility.
It answers exact lookups in both directions (what a table contains, and which tables carry an item), searches by partial description or pattern when the name is unknown, and ships with several levels of helpful guiding for the target LLM so answers are read the JDE way. No database connection at runtime, no SQL in the AI’s path, immediate responses.
MCP server implementation guides
See how the individual MCP servers expose JDE Orchestrations, controlled file access, distributed logs or the JDE data model as narrow, named tools rather than an open-ended integration surface.
Need to expose an existing STDIO MCP server to Copilot?
Steward is the secure MCP gateway for Microsoft Copilot and compatible remote enterprise AI clients. Existing tool-oriented STDIO MCP servers can normally remain unchanged while access to configured protected resources stays bound to the requesting user’s actual identity and existing permissions.
See Steward →Licensing
Annual subscription, counted per computer on our licensing server — licence count and expiry are controlled centrally. Self-serve buying currently applies to the JDE Orchestration MCP server only. Order or inquire, or see the pricing page for current pricing.
Background reading
Plain-English context on this product’s space — what it is, why it matters, and how it fits into a controlled JDE automation strategy. These guides live on JDE Citizen Developer.