How to use Claude with DELMIAworks (IQMS)
Short answer
Claude reaches DELMIAworks through a connector you build, typically an MCP server that wraps read-only database views or the vendor API. You can run Claude through Anthropic directly, AWS Bedrock or Google Vertex AI, and plants with restricted data can swap in an on-prem open model behind the same tools.
Manufacturers on DELMIAworks, formerly IQMS, increasingly want Claude to read an order, explain a work order delay or draft a corrective action report. Claude is strong at long documents and careful reasoning, which suits quality records and contract review as much as ERP lookups.
What Claude lacks is a built-in link to your Oracle-backed ERP. The bridge is a connector, and the Model Context Protocol (MCP) is now the common way to build one so it works across Claude apps and other clients.
This page covers the hosting choices, what to expose, and how data handling differs between the Anthropic API, cloud marketplaces and an on-prem setup.
What teams use it for
Corrective action drafting
Claude reads the nonconformance record, the lot genealogy and the work order history, then drafts an 8D style report for the quality engineer to edit.
Customer contract review against orders
Paste a customer quality agreement and let Claude flag order lines or lead times that conflict with its terms.
Plain-language order investigation
A planner asks why a mold change ran long, and Claude reads press events and the setup routing to answer.
Engineering change summaries
Claude compares two BOM revisions and writes a summary of what changed and which open orders are affected.
Long report digestion
Feed in a quarter of scrap and downtime exports and ask for recurring patterns by mold and material lot.
Onboarding Q&A
New schedulers ask how the plant handles regrind or color changes, answered from internal SOPs plus live data.
DELMIAworks (IQMS) integration surface
- Oracle read views: curated views over items, orders, purchase orders, work orders and MES events give Claude a safe, fast data source.
- Vendor web services: where your DELMIAworks release offers REST endpoints, an MCP server can call them with a service account and honor application roles.
- Attachments and documents: quality certificates, drawings and procedures are often stored as files or linked documents, which Claude can read when exposed through a document tool.
- Reporting layer: saved reports provide stable, named datasets that are easy to expose as tools.
- Version note: object names and endpoints differ between releases, so validate with your administrator.
How it works
- 1
Choose the Claude route
Use Claude directly through Anthropic, or through AWS Bedrock or Google Vertex AI if your organization already has cloud agreements and regional controls there.
- 2
Write an MCP server
Create a small server in Python or TypeScript that exposes named tools such as get_order, get_workorder_events and search_items.
- 3
Back tools with views
Each tool runs a fixed parameterized query on a read-only Oracle account or calls a vendor endpoint.
- 4
Pick a transport
Use stdio for a developer's local testing and HTTP with authentication for shared team use.
- 5
Map identity
Tie each user to a role so Claude only sees the plants, customers and fields that person may see.
- 6
Test with real questions
Collect twenty questions planners ask and verify each answer against the source screens.
- 7
Add documents carefully
Index quality procedures separately and attach them as a retrieval tool with source citations.
Example flows
Draft an 8D from a complaint
Trigger: Customer reports flash on a molded housing.
- Quality engineer gives Claude the lot number.
- Claude calls the genealogy tool for resin lot, press, mold and shift.
- It pulls related MES events and prior scrap for that mold.
- It drafts containment, probable causes and next checks with sources.
Result: The engineer starts from a structured draft rather than a blank form.
BOM revision impact
Trigger: Engineering releases revision C of an assembly.
- Claude reads both revisions through the BOM tool.
- It lists added, removed and changed components.
- It checks open work orders and POs that use affected items.
- It writes a short impact note for the planner.
Result: Planners see which open jobs need attention before revision C reaches the floor.
Keeping data on-prem
When data must stay inside the plant, Claude's hosted options may not fit, since the model runs in Anthropic or cloud provider infrastructure. Bedrock and Vertex let you keep processing within your cloud account and region, which satisfies many policies but is still off-site. For air-gapped or export-controlled sites, keep the same MCP server and tool design, and point it at an open-weight model served locally with vLLM or Ollama. The tools, permissions and logs stay identical, and you can use Claude for non-restricted workloads while the private model handles restricted ones.
Security checklist
- Review Anthropic's commercial terms and any cloud provider terms for retention and training commitments before sending ERP data.
- Run the MCP server with a read-only Oracle account and parameterized queries only.
- Authenticate HTTP transport and map each caller to an ERP role.
- Log every tool call with user, arguments and result size.
- Exclude export-controlled and customer-restricted records from any hosted model path.
Ways to get it done
Existing connectors or community servers
Effort: Days, but review the code and permissions carefully.
Check whether a DELMIAworks or generic Oracle MCP server meets your needs before building.
Build your own MCP server
Effort: Two to four weeks for a solid read-only set.
Best control over fields, roles and logging for teams with developer capacity.
Partner such as Netray
Effort: A few weeks to pilot.
When you need Claude for some work and an on-prem model for the rest, with one shared tool layer.
Pitfalls
- !Giving the MCP server the application owner database account.
- !Exposing free-form SQL as a tool and letting the model improvise.
- !Sending restricted data through a hosted route because the demo was easier.
- !Skipping source citations, so users cannot tell which table an answer came from.
- !Ignoring context size: dumping a whole quarter of events instead of summarizing server side.
Frequently asked questions
Does Claude have a native DELMIAworks connector?
Do not assume one exists. Claude supports connectors built on MCP, and you can build one for DELMIAworks over read-only views or vendor endpoints. Check current marketplaces, then review any third-party server before trusting it with plant data.
Is Claude on Bedrock or Vertex different?
The model capability is similar, but processing happens inside your cloud account and region with that provider's controls and billing. Many enterprises prefer this route for governance. It is still a hosted service, not on-premises.
Can Claude run fully on-prem?
Claude models are served by Anthropic and its cloud partners, so a fully air-gapped deployment uses an open-weight model instead. You can keep the same MCP tools so switching models does not mean rebuilding the integration.
What should the first tools be?
Start with read-only item lookup, order status, work order status and open PO lookup. These are low risk, easy to verify and cover many daily questions. Add genealogy and BOM tools next.
How do I stop Claude seeing everything?
Scope at the tool layer, not in the prompt. Use a restricted database account, filter plants and customers in the query, and map each user to a role. The model cannot request what the tool will not return.
Want Claude working on your DELMIAworks (IQMS)?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment