How to use Claude with ECI M1
Short answer
The clean route is a Model Context Protocol server in front of M1's SQL Server data, which Claude connects to as a tool source. Where cloud use is allowed, Claude is available through Anthropic's own offerings and through AWS Bedrock or Google Vertex AI. Where it is not, the same tool layer can drive an on-prem open-weight model.
ECI M1 teams usually ask about Claude for two reasons: strong reading and reasoning over long documents such as quotes, specs and supplier terms, and the Model Context Protocol, which gives a standard way to attach tools to an assistant instead of pasting data.
The question is not whether Claude can talk to M1, but through what. M1 has no widely published Claude connector that we would rely on, so the integration is something you build or commission: a small server that exposes a few well-scoped read tools over the M1 data.
The other half of the decision is where Claude runs. That is a data-governance question, and for many job shops serving aerospace and defense customers, it decides the architecture before any code is written.
What teams use it for
Long-document review beside live data
Ask Claude to read a customer purchase order's terms and compare delivery and material requirements with what M1 shows for the matching sales order.
Supplier quote comparison
Feed several vendor quote PDFs and have Claude line them up against the item's last PO price and lead time pulled through a tool.
Planner Q&A
Ask which jobs are blocked on a given component and what supply is expected, answered through tool calls rather than memory.
Engineering change impact notes
Given an ECO description, Claude lists the open jobs and POs touching the affected part so a planner can assess impact.
Audit evidence packs
Assemble a narrative with citations to the specific job, lot and receipt records for a customer or quality audit, reviewed by a person.
ECI M1 integration surface
- SQL Server database behind M1, read through a documented view layer.
- Any supported M1 web services or import path (verify by version and licence) for approved writes.
- A custom MCP server that maps view queries to named tools; this is the connector Claude actually uses.
- Document stores holding drawings, certificates and POs, often a file share or a document management add-on, which Claude can read when exposed through a tool.
- Your identity provider (Active Directory or Entra ID) for authenticating users to the MCP server.
How it works
- 1
Decide where Claude may run
Options include Anthropic's enterprise offerings, Claude on AWS Bedrock or Google Vertex AI in a region you choose, or an on-prem model instead. Settle this with security and your customers' flow-down clauses first.
- 2
Create the M1 view layer
Build views for items, orders, jobs, POs and inventory with business-friendly names, on a replica where possible.
- 3
Write the MCP server
Expose a few tools with typed parameters, row limits and clear descriptions. Start with item lookup, order status and open PO by vendor.
- 4
Connect Claude
Register the server with the Claude client your team uses, or call it from your own application via the API. Keep the server on your network and reachable only through your access path.
- 5
Scope by user
Pass the user's identity through so the server applies their M1 role when filtering results.
- 6
Evaluate on real questions
Run a test set with known answers, review tool-call traces, and tune tool descriptions before rolling out.
Example flows
PO terms against the sales order
Trigger: A contract reviewer uploads a customer PO PDF and names the M1 sales order.
- Claude reads the PDF for quantities, dates and special requirements.
- It calls the order tool for the matching M1 order lines.
- It lists mismatches such as quantity, price or due date, and unmentioned certification clauses.
- The reviewer corrects M1 or contacts the customer.
Result: Discrepancies are caught at order entry rather than at shipment.
Component shortage briefing
Trigger: A planner asks which open jobs depend on a constrained alloy.
- Claude calls the where-used and open demand tool for the item.
- It calls the supply tool for on-hand and open PO lines.
- It writes a short briefing sorted by job due date.
Result: The planner has a ranked list to act on in minutes.
Keeping data on-prem
Claude models are not distributed for you to run on your own servers, so a strictly air-gapped M1 site cannot use Claude itself. What transfers is the architecture. Build the MCP server against M1 once, then point it at a cloud Claude where policy allows, and at an open-weight model served locally with vLLM or Ollama where it does not. Many MCP-capable clients and agent frameworks work with local models, so the tools do not change. For a hybrid, route by data class: non-sensitive drafting to a cloud Claude, anything touching controlled parts or customer drawings to the local model.
Security checklist
- Review the data-handling, retention and regional terms of the specific Claude offering you choose against customer contracts.
- Authenticate users to the MCP server and apply M1 role filters on every tool.
- Keep tools read-only initially; add approval-gated writes later.
- Use a least-privilege SQL login for the server, scoped to views.
- Log all tool calls and keep them for audit and troubleshooting.
Ways to get it done
Claude with a vendor-neutral MCP server you build
Effort: Several weeks for a first useful toolset.
Best if you have SQL skills in-house and clear cloud permission.
Claude via AWS Bedrock or Google Vertex AI in your cloud tenancy
Effort: Adds cloud setup on top of the MCP build.
When your organisation already standardises on one of those clouds and wants its network and logging controls.
Partner-delivered build with an on-prem fallback
Effort: A scoped pilot, usually weeks.
When some data cannot leave the plant.
Pitfalls
- !Assuming a ready-made M1 connector exists and finding out late that you must build one.
- !Exposing too many tools at once, which makes tool selection less reliable.
- !Treating the model as the security boundary instead of the MCP server.
- !Sending whole drawings or cost sheets to a cloud service without checking contract flow-downs.
- !No row limits, so a vague question pulls a huge table into context.
Frequently asked questions
Is there an official Claude connector for ECI M1?
We are not aware of one you should depend on. The practical approach is a custom MCP server over your M1 data. Check with ECI and your reseller in case a connector has been released for your version.
What is MCP and why use it?
Model Context Protocol is an open standard for connecting assistants to tools and data. Writing one MCP server for M1 lets Claude and other MCP-capable clients use the same tools without separate integrations.
Can Claude run fully on-prem with M1?
No, Claude itself is a hosted model. For fully on-prem work use an open-weight model and keep the same MCP tool layer. You can still use Claude for non-sensitive tasks.
Does Claude learn from our M1 data?
That depends on the plan and terms you choose. Enterprise and cloud-provider routes generally offer controls around training and retention. Read the current terms for your chosen route and have legal confirm.
Can Claude change records in M1?
It can if you build write tools, but start read-only. If you add writes, route them through approval and a supported M1 import or API path.
Want Claude working on your ECI M1?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment