# How to connect Claude to MRPeasy

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

## Short answer

Claude reaches MRPeasy through a Model Context Protocol server that wraps the MRPeasy REST API, giving it read-only tools for orders, stock and BOMs. You can run Claude via Anthropic directly or via a cloud such as AWS Bedrock or Google Vertex, and use an open-weight model when data must stay on your own hardware.

## System

MRPeasy

## Capability

claude

## Intro

Claude is useful for the reading-heavy parts of running an MRPeasy shop: long supplier emails, drawings notes, customer specifications, and the question of what an order status really means. The model is good at turning a messy question into the right sequence of lookups and then explaining the answer.

It cannot see MRPeasy by itself. The cleanest bridge is a Model Context Protocol server that holds the API credentials and exposes a few named tools. Claude clients that support MCP can then call those tools, and you decide exactly which ones exist.

Because MRPeasy is hosted, there is no database to attach. The MCP server speaks to the REST API, and your main design choices are where that server runs and where the model runs.

## Use Cases

### Order status in plain language

**Description:** Ask Claude what is holding up an order. It calls the order and manufacturing tools and summarises the blockers with dates.

### Supplier email triage

**Description:** Paste a supplier reply about a delay and have Claude match it to the open purchase order and draft the production impact note.

### Spec review against BOM

**Description:** Give Claude a customer specification and the BOM tool output, and ask which components may conflict with a material or finish requirement.

### Costing explanations

**Description:** Ask why a product's cost changed between two months and have Claude compare the BOM and purchase prices it retrieves.

### Onboarding new planners

**Description:** Let a new hire ask how an item flows from purchase to stock to manufacturing order, answered from your own data and naming.

### Weekly management summary

**Description:** Generate a short narrative of orders shipped, late jobs and top shortages, with each claim tied to a record.

## System Surface

- REST API authenticated by API key and secret, the foundation of any MCP server. Check your account's API documentation for current endpoints and field names.
- Customer orders, manufacturing orders, purchase orders, items, BOMs and stock are the objects most worth exposing as tools.
- API rate limits: batch and cache inside the MCP server so a single Claude question does not trigger dozens of calls.
- Report exports to CSV or Excel, useful for attaching snapshot files when you do not want live access.
- No on-premise database. A scheduled sync into your own store is the way to give Claude fast, joined data without touching the vendor API per question.

## How It Works

### Choose the Claude surface

**Detail:** Claude desktop or web with connectors for a few users, or the API for an internal app. Confirm which plan supports custom connectors.

### Build the MCP server

**Detail:** Write a small server in TypeScript or Python that registers tools like get_order_status, get_item_stock and get_bom, each calling the MRPeasy API.

### Pick a transport

**Detail:** Use stdio for a single local user, or HTTP with authentication when several people share it from a server.

### Constrain results

**Detail:** Return compact JSON with only needed fields and a record count, so the model is not flooded with whole order histories.

### Add instructions and guardrails

**Detail:** Tell Claude to cite record numbers, state when a lookup returned nothing, and never infer dates.

### Evaluate before rollout

**Detail:** Run 25 to 50 real questions and compare answers to the MRPeasy screens, then widen access.

## Example Flows

### Supplier delay to production impact

**Trigger:** A buyer pastes a supplier email saying a part ships two weeks late.

**Steps:**

- Claude identifies the part number and calls the purchase order tool.
- It looks up manufacturing orders consuming that part with the order tool.
- It lists which customer orders are affected and by how many days.
- It drafts a note for the planner with the options.

**Result:** The planner sees the impact in one reply instead of tracing it through three screens.

### Customer spec check

**Trigger:** Sales receives a customer drawing note requiring a specific material.

**Steps:**

- Claude reads the note and calls the BOM tool for the quoted product.
- It compares listed materials and flags any mismatch.
- It suggests questions for engineering.
- The salesperson attaches the answers to the quote.

**Result:** Fewer quotes go out with a material the shop cannot supply.

## On Prem

There are three hosting levels. First, Claude through Anthropic's own API and apps. Second, Claude through a cloud platform such as AWS Bedrock or Google Vertex AI, which lets you use your existing cloud contract, regions and network controls. Third, no Claude at all: an open-weight model on your own servers, using the same MCP server. MRPeasy remains in the vendor's cloud in every case, so the question is where prompts and results go. If a customer contract, export regulation or classified program says no external model, take the third route and keep the MCP server on your network. The tool design carries over unchanged, which is why it is worth building the MCP layer once and cleanly.

## Security

- Hold the MRPeasy API secret only inside the MCP server, never in prompts or client config files shared among users.
- Expose read-only tools first and review each new tool as a code change.
- Authenticate users to the MCP server over HTTP and scope tools by role.
- Log tool calls with user, arguments and result size for audit.
- Review the data terms of your chosen Claude plan or cloud platform, including retention and region.

## Options

### Community or third-party MRPeasy MCP server

**Fit:** Quick trials, if one exists and you can review its code and permissions.

**Effort:** Hours, plus a security review

### Build your own MCP server

**Fit:** Teams with a developer, wanting control over tools, fields and logging.

**Effort:** 1-3 weeks

### Partner build with Netray

**Fit:** When you need role scoping, evaluation sets, and an optional on-prem model alternative.

**Effort:** Typically a few weeks for a scoped pilot

## Pitfalls

- Exposing a generic fetch-any-URL tool. Give the model named, bounded tools instead.
- Returning entire records. Large payloads waste context and increase wrong answers.
- Sharing one API key across all users with no per-user trace.
- Assuming date fields are all in one timezone or format. Normalise in the server.
- Skipping evaluation, then discovering the model summarised a status the way you did not intend.

## Faq

### Does Claude have a native MRPeasy integration?

**Answer:** There is no first-party one that we can point to with confidence, and availability changes. The dependable route is a custom MCP server wrapping the REST API, which any MCP-capable Claude client can use.

### What is MCP and why use it?

**Answer:** Model Context Protocol is an open standard for giving AI clients tools and data. You write one server for MRPeasy and it works with Claude and other MCP clients, instead of building a separate connector for each product.

### Can Claude run inside our network?

**Answer:** Claude models are served by Anthropic and by cloud partners, not installed on your hardware. For fully on-prem use, run an open-weight model and reuse the same MCP server and tool definitions.

### Is Bedrock or Vertex better than direct access?

**Answer:** They are better if your company already standardises on that cloud, needs its regional controls, or wants billing and access policy in one place. Direct access is simpler for a small team that just wants to start.

### Can Claude update orders in MRPeasy?

**Answer:** Only if you build write tools. We recommend delivering read-only first, then adding one carefully reviewed write such as a draft record with a human confirmation, and logging every call.

## Search terms

mrpeasy claude, claude for mrpeasy, connect claude to mrpeasy, mrpeasy anthropic, mrpeasy claude mcp, claude erp integration mrpeasy, mrpeasy bedrock claude, mrpeasy claude api

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