# How to connect Claude to Made2Manage

> Source: https://netray.co/integrations/made2manage-claude
> Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems.
> Section: Integrations

## Short answer

Claude reaches Made2Manage through a connector you provide, most commonly an MCP server that wraps a read-only view of the SQL Server database. You can run Claude through the Anthropic API or a cloud platform like AWS Bedrock or Google Vertex, or use an open-weight model on-prem when data must not leave.

## System

Made2Manage

## Capability

claude

## Intro

Made2Manage shops tend to have strong tribal knowledge and weak search. The data to answer 'what do we owe this supplier' or 'which jobs touched this lot' exists, but it takes several screens and a long-tenured employee. Claude is good at reading structured results, reasoning about them, and writing a clear answer.

The gap is access. Claude does not know your database. The modern bridge is the Model Context Protocol: you expose a few tools, Claude calls them during a conversation, and you control exactly what comes back.

Where Claude runs matters as much as how it connects, especially for an on-prem ERP with controlled customers. This page covers the connection options and the hosting choices side by side.

## Use Cases

### Ask questions across jobs and POs

**Description:** A planner asks which open jobs are blocked by a part with a late PO. Claude calls a shortage tool and an open-PO tool, then explains the chain.

### Review a routing before release

**Description:** Claude reads a new job's routing and bill of material and flags missing operations or outside-service steps compared with similar past jobs.

### Long document reading next to ERP data

**Description:** Supply a customer PO or quality clause PDF and let Claude compare it with the sales order lines retrieved from the ERP.

### Month-end explanations

**Description:** Claude drafts commentary on inventory and WIP movements from summarised ledger and job data supplied by tools.

### Onboarding new staff

**Description:** New buyers ask how a process works at your company and Claude answers from your procedures document set plus live lookups.

## System Surface

- Claude has no native Made2Manage connector; the usual bridge is your own MCP server in front of SQL Server.
- A read-only database login and parameterised queries are the foundation. Table names vary by version, so map them from your schema.
- Reports and exports can feed Claude as files for one-off work, with the usual staleness limits.
- Aptean may provide integration APIs depending on your release and contract; verify before building a custom data layer.
- Because many installs sit behind a firewall, the MCP server often runs inside the network and Claude reaches it through a controlled gateway or a local desktop client.

## How It Works

### Pick the Claude access route

**Detail:** Options include the Claude apps with connectors, the Anthropic API, or Claude through AWS Bedrock or Google Vertex AI if your company already standardises on those clouds.

### Build the MCP server

**Detail:** Implement tools such as get_sales_order, get_job_status, open_po_by_part, stock_by_part. Return compact JSON with units and as-of timestamps.

### Connect transport and auth

**Detail:** Use stdio for a single user's desktop client, or HTTP behind your gateway with tokens for a shared service.

### Write tool descriptions carefully

**Detail:** Clear names and descriptions matter more than clever code, because Claude chooses tools by reading them.

### Test with real questions

**Detail:** Collect twenty questions your planners really ask and check answers against the screens.

### Roll out by role

**Detail:** Give buyers, planners and sales different tool sets so nobody sees data outside their job.

## Example Flows

### Planner asks about a stuck job

**Trigger:** Question in Claude: why is job 12345 not moving?

**Steps:**

- Claude calls get_job_status and reads the current operation and last posting.
- It calls shortages_for_job and finds a missing component.
- It calls open_po_by_part and sees a promise date two weeks out.

**Result:** A one-paragraph explanation plus a suggested action, no screen hopping.

### Contract review against sales order

**Trigger:** User attaches a customer PO PDF.

**Steps:**

- Claude extracts lines, quantities and dates from the PDF.
- It retrieves the matching sales order through a tool.
- It lists mismatches in price, quantity or delivery.

**Result:** Fewer order-entry errors caught before the job is released.

## On Prem

If shop data cannot go to a hosted model, keep the MCP-style tool layer and swap the model. An open-weight model such as a Qwen, Llama or Mistral class model, served on a local GPU with vLLM or Ollama, can call the same tools through an MCP-compatible client. Everything stays on your network, including prompts and results. The honest trade-off is that smaller local models are weaker at long, multi-step reasoning than the largest hosted models, so test your real questions before committing. Some organisations accept a hybrid: private model for sensitive lookups, hosted Claude for non-sensitive drafting.

## Security

- Review the provider's enterprise data terms for retention, training use and regional processing before connecting live data.
- Run the MCP server with a read-only database account and parameterised queries only.
- Authenticate remote MCP access and avoid exposing the ERP network to the internet.
- Log each tool call and who invoked it.
- Do not send export-controlled technical data to a hosted model unless your compliance team has approved the arrangement.

## Options

### Claude app plus a self-built MCP server

**Fit:** Small team with a developer and non-sensitive data.

**Effort:** A few weeks for a useful read-only toolset.

### Claude via Bedrock or Vertex in your cloud account

**Fit:** You need cloud controls, regional hosting and existing cloud governance.

**Effort:** Several weeks including cloud and network setup.

### On-prem open model with a partner such as Netray

**Fit:** Sensitive or controlled data; air-gapped or restricted networks.

**Effort:** Scoped pilot over several weeks, then expansion.

## Pitfalls

- Exposing too many tools; Claude then picks the wrong one more often.
- Returning huge result sets that waste context and slow answers.
- Missing as-of timestamps, so users think a replica is live.
- Skipping role scoping because the first users are administrators.
- Assuming a hosted model is approved for customer-restricted data without checking contracts.

## Faq

### Does Claude have a Made2Manage connector?

**Answer:** Not that I would rely on. Claude connects through MCP servers that someone builds, so for Made2Manage you write or commission a server over a read-only database view. Check with Aptean whether any official integration exists for your release.

### Can Claude run inside our network?

**Answer:** Hosted Claude models are not self-hosted on your own hardware. You can run the connection layer inside your network and reach Claude via the API or a cloud platform. For fully local inference you would use an open-weight model instead.

### What is the benefit of Bedrock or Vertex?

**Answer:** They let you use Claude under your existing cloud agreements, with your cloud's identity, logging and regional controls. That can simplify procurement and governance, though your ERP data still travels to that cloud when prompts are sent.

### Will Claude change records in Made2Manage?

**Answer:** Only if you give it write tools, which we advise against at first. Start read-only, have it draft actions, and let a person enter or approve them through the normal application so ERP business rules still apply.

### How do we keep answers accurate?

**Answer:** Return structured data from tools, include as-of times, and require the answer to cite the fields it used. Test against known questions regularly, and fix data quality problems, since the model will faithfully summarise bad records.

## Search terms

made2manage claude, claude for made2manage, made2manage claude mcp, connect claude to made2manage, made2manage anthropic, claude erp on-prem manufacturing, made2manage bedrock claude, made2manage ai assistant

---
Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json