How to add AI agents to MRPeasy
Short answer
MRPeasy is a cloud MRP with a REST API, so an AI agent is a small service that reads orders, stock and BOMs through that API, reasons over them, and proposes actions a planner approves. Start read-only with shortage triage, then add narrow, approved writes such as draft purchase orders.
MRPeasy teams are usually small manufacturers with one or two planners who live in the Purchasing, Manufacturing Orders and Stock screens. The pain is rarely missing data. It is the daily grind of noticing which customer order is about to miss its date because a purchased part is late, and then chasing it by email.
An AI agent is not a chatbot that answers questions. It is a worker with a goal, a set of tools that call the MRPeasy API, and a rule that it may only propose changes. It wakes on a schedule, pulls what changed, decides what matters, and hands the planner a short list with the evidence attached.
MRPeasy is delivered as SaaS, so there is no database to point an agent at. Everything goes through the documented API or through scheduled exports. That is a constraint, but it also makes the integration clean: one authenticated surface, one audit trail.
What teams use it for
Late-material triage
Each morning the agent compares open manufacturing order components against purchase order due dates and stock, then ranks which jobs are at risk and why.
PO expediting drafts
For purchase orders past due or due this week, the agent drafts a supplier chaser with the part, quantity and promised date, ready for the buyer to send.
Reorder proposals
The agent reads stock levels, open demand and lead times and proposes draft purchase orders for items below their real need, not just below a static minimum.
Quote drafting
From a customer RFQ email, the agent finds the product and BOM, rolls up material cost and routing time, and drafts a quote for review.
Order promise checks
Before a salesperson commits a date, the agent checks capacity and material and returns a defensible delivery estimate with assumptions listed.
Data hygiene sweeps
The agent flags items with zero lead time, BOMs missing routing steps, or vendors without prices, which quietly break MRP output.
MRPeasy integration surface
- REST API (typically under app.mrpeasy.com/api, version path varies) with endpoints for items, products, BOMs, customer orders, manufacturing orders, purchase orders, vendors, customers, shipments and stock. Confirm exact paths in your account's API documentation.
- Authentication with an API key and secret generated by an administrator in the integration settings. Treat the pair as a service credential with its own access rights.
- Request rate limits apply on the API. Plan for pagination and back-off rather than pulling the whole item list on every run.
- Webhook or event notification support depends on plan and version. If it is not available, schedule polling on modified-since filters.
- Report and list exports to CSV or Excel from the UI, useful as a fallback snapshot for backfills and for checking that the agent's API view matches what planners see.
How it works
- 1
Create a dedicated API user and key
Generate the key and secret under a role that can read what the agent needs. Do not reuse an administrator key.
- 2
Build a thin tool layer
Wrap the API in a handful of typed functions such as get_open_pos, get_mo_components, get_stock and get_bom. The model never sees raw URLs or the secret.
- 3
Cache and normalise
Pull changes on a schedule into a small local store so the agent can join orders, stock and POs without hammering the rate limit.
- 4
Give the agent a narrow goal and rules
For example: find manufacturing orders due within 10 days whose components will not arrive in time. State what counts as at risk and what it may never do.
- 5
Return proposals, not writes
The output is a ranked list with the order numbers, shortage quantities and a suggested action. Writes happen only after a planner clicks approve.
- 6
Log everything
Store each run's inputs, tool calls, reasoning summary and the approver so you can replay a decision later.
Example flows
Morning shortage digest
Trigger: A scheduled job at 06:30 local time.
- Fetch manufacturing orders with start dates in the next 10 days.
- Fetch component requirements and compare to on-hand stock and open PO due dates.
- Rank orders by customer due date and by how many components are short.
- Draft one line per risk with the cause and a suggested action.
Result: The planner opens one digest instead of walking five screens, and starts with the orders that will actually slip.
Supplier chaser draft
Trigger: A purchase order passes its expected receipt date without a receipt.
- Look up the PO lines, vendor contact and the manufacturing orders waiting on the part.
- Draft a short email stating the part, quantity and the date production needs it.
- Queue the draft for the buyer to review and send.
- If the buyer records a new date, update the draft note for the next digest.
Result: Expediting happens the same day the date is missed, with the production impact quoted.
Keeping data on-prem
MRPeasy itself is cloud-hosted, so there is nothing to install on your own server for the ERP. What you can keep private is everything around it: the agent runtime, the local cache and the language model. A small on-prem or private-cloud box can poll the MRPeasy API over HTTPS, store a copy of the working data, and run an open-weight model locally so supplier names, pricing and customer orders are not sent to a third-party AI provider. The only outbound traffic to the cloud is the API calls to MRPeasy. This suits defense and aerospace subcontractors who must control where order data is processed.
Security checklist
- Use a dedicated API key with the minimum read scope, and rotate it on a schedule.
- Keep the agent read-only at first. Add writes one narrow action at a time behind approval.
- Store the key in a secrets manager, never in prompts or code.
- Log every tool call and approval with user, time and payload.
- If you use a hosted model, send only the fields the task needs and review the provider's data retention terms.
Ways to get it done
Automation tools such as Zapier or Make with an AI step
Effort: Days
Simple single-trigger tasks like a Slack alert on a late PO. Weak for multi-step reasoning across orders.
Build in-house with the API and an LLM framework
Effort: 3-8 weeks for a first useful agent
Teams with a developer who can maintain a small service and an evaluation set.
Partner build with Netray
Effort: Typically a few weeks for a scoped pilot
When you want a private model, approval workflow and audit trail set up and tuned to your data.
Pitfalls
- !Letting the agent write purchase orders directly. Start with drafts and a human click.
- !Ignoring rate limits and getting blocked mid-run. Cache and paginate.
- !Trusting lead times in the item master. Bad master data produces confident, wrong advice, so run a hygiene sweep first.
- !Measuring success by demo quality. Keep a list of past late-material cases and check the agent's calls against what really happened.
- !Giving the agent one huge prompt with every table. Small tools and a narrow goal are more reliable.
Frequently asked questions
Can an AI agent change data in MRPeasy?
Technically yes, if the API key allows writes. In practice you should start with read-only access, then enable a single write such as creating a draft purchase order, always behind human approval. Keep a log of who approved each action.
Do I need MRPeasy's permission to build this?
You use the API your subscription provides, so no special partnership is needed. Check your plan for API access and any rate limits, and review the terms of service for automated access before going to production.
Which model should the agent use?
For tool-calling over structured data, a strong hosted model works well, and a mid-size open-weight model can work when run locally with a tight tool set. Test both against your own historical cases before choosing.
How is this different from the ERP's built-in planning?
MRPeasy plans from your rules. An agent adds judgement on top: it notices patterns across suppliers, drafts the chasers, and explains the risk in words. If your planning needs are simple, the built-in tools may be enough.
How long until it is useful?
A read-only shortage digest can run within a couple of weeks once the API key and data quality checks are in place. Approved write actions usually follow after the team trusts the digest.
Want AI agents working on your MRPeasy?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment