Magic Integration
Platform, IT Services & AI Development

Not one more system to manage. The system that manages everything else.

Magic Integration is the enterprise integration platform, the applications we build on top of it, the AI engineering that makes those systems answerable, and the managed IT services that keep all of it running after go-live. One division of Magic Software Americas. One P&L. Nothing subcontracted.

The Scope Problem

Most buyers think this is ERP integration. That undersells it by about seventy percent.

When someone says integration, they usually picture one pipe between an ERP and one other thing. That is a project. What you actually own is an estate — every place a number leaves one system and has to arrive, correct and on time, in another. Magic XPI is an enterprise integration platform, and the word enterprise is doing real work in that sentence.

ERP

Orders, items, costs, inventory, GL. The system of record that every other system wants to read from and almost none of them are allowed to write to.

CRM

Quotes and accounts that have to become real orders without a person retyping them. Salesforce, Dynamics, SugarCRM.

WMS

Pick, pack, ship and cycle counts. The warehouse knows what is on the shelf hours before the ERP does.

MES

Work orders down, completions and scrap back up. The layer that decides whether your routings reflect reality.

HR

Hires, moves and exits that should provision an account, a badge and a training record without three tickets.

Finance

AP, AR, tax, banking and consolidation. Close cycles that stall because a subledger never got the file.

Support

Cases, RMAs and field service calls that need entitlement and shipment history to be answerable in one screen.

Historians

Tags, trends and time-series. Decades of process truth that finance and planning have never once queried.

Plant floor

Counts, states, downtime reasons and quality results, moving up through OPC UA and MQTT instead of a clipboard.

How the estate is actually held together today

  • A flat file on a share that one scheduled task picks up at 2 a.m., if it runs.
  • A script written by someone who left, in a language nobody left here writes.
  • A point-to-point connector sold with the application, which only speaks to that application.
  • A person with a second monitor, reading one system and typing into another.

What changes when the estate has one owner

  • Every flow is defined in one place, versioned, and visible when it fails.
  • A new system is a connector and a mapping, not a new integration project.
  • Retries, alerts and recovery are platform behavior, not something each script reinvents.
  • The people who were rekeying orders go back to the work you hired them for.

Count your own handoffs before you scope anything. Every one of them is a connector, a script somebody left behind, or a person retyping. Only the first of those three has an owner.

The Platform

Four products. One integration foundation underneath all of them.

We build and own this software. That matters when something breaks at the layer nobody else will claim, because the engineer who can fix it works here.

Integration Platform

Magic XPI

Enterprise integration across ERP, CRM, WMS, MES, HR, finance, support, historians and the plant floor. Code-free flow design, event-driven execution, 100+ pre-built and certified connectors, and one console where every flow either ran or told you why it did not.

  • Visual mapping across XML, JSON, CSV, EDI and relational data
  • Event-driven and batch in the same platform
  • Cloud, on-prem or hybrid deployment
Explore Magic XPI

Application Development

Magic SmartUX

The application layer on top of the systems you already run. Magic SmartUX builds mobile and web experiences against existing enterprise logic, so the warehouse, the field tech and the plant supervisor each get a screen that fits the job instead of an ERP form that fits nobody.

  • Give us a screenshot, we give you an app
  • Build once, deploy anywhere — web, iOS, Android
  • Existing business logic reused, not rewritten
Explore Magic SmartUX

Finance & Plant Apps

Magic FactoryEye

Four apps on one governed data layer: Finance Eye for order and invoice capture, Plant Finance for real-time job cost, and Plant Analytics and Plant Modeling for floor visibility. Every app reads through Data Hub on Magic XPI rather than wiring itself into your control systems. We are honest about where it sits: the plant apps are early, with one production customer, and we scope them on an integration foundation.

  • Reads via Magic XPI — not a direct tap on MES or SCADA
  • Start with one app; the next reuses the same connection
  • Plant apps early, one production customer to date
Explore Magic FactoryEye

AI Governance Layer

Magic MCP

A model-agnostic governance layer between your language models and your operational systems. It decides what a model may see, what it may do, and what gets written down about it. Three deployment modes: your cloud with controls, a private VPC, or air-gapped on-prem.

  • Model-agnostic — swap models without rebuilding access
  • Scoped, audited access to live system data
  • Your cloud, private VPC, or air-gapped on-prem
Explore Magic MCP
Architecture

A platform, not a consultancy with a toolkit. Here is the difference, in detail.

If you are the person who has to decide whether this is real software or a services shop with a favorite framework, this section is for you. Magic has been building integration software since 1983. Magic XPI is our product, on our own middleware, with our own connector catalog.

Magic XPI at the center of an ecosystem of ERP, CRM, databases, APIs, messaging and file systems

100+ pre-built and certified connectors

Certified means tested against the vendor's own interfaces and maintained through their release cycles — not a generic REST call pointed at an endpoint and called an integration. When SAP, JD Edwards or Epicor change, the connector is our problem to keep current, not yours.

Code-free flow design

Flows are built visually: triggers, steps, branches, mappings, error paths. That is not a convenience feature, it is a maintenance strategy. Three years from now the flow is still readable by whoever is on call, instead of being a binary nobody will touch.

Event-driven, not just scheduled

A flow can fire on a record change, a queue message, a file arriving, an MQTT topic or an API call, as well as on a schedule. Batch windows exist because the tooling needed them. Yours should exist because the business wants them.

Proprietary in-memory middleware

Flow state lives in our own in-memory engine rather than being parked in a database between every hop. That is what gives the platform its throughput profile, its high-availability behavior and its ability to resume a partially completed transaction instead of replaying it and double-posting.

High availability as a configuration

Multiple servers share the same flow definitions and the same state, so a node going down is a failover, not an outage with a manual cleanup afterward. Monitoring, alerting and retry policy are platform-level, which means every flow inherits them.

Cloud, on-prem or hybrid

The same platform runs in your cloud tenant, in your data center, or split across both — which is the common case, because the ERP is hosted and the historian is never leaving the plant. Governance and data residency decide the topology; the flows do not change.

The evaluator's shortcut

Read the connector list. It is the fastest way to tell a product from a promise.

View the Connector Library
Systems We Connect

We do not ask you to replace the system of record. We make it reachable.

Deep, specific experience in the platforms that run mid-market and enterprise manufacturing and distribution. Not a list of logos — a set of connectors, data models and upgrade scars.

The Strike Team

The integrations that stall a project are rarely the ones in the catalog. They are the two systems nobody owns.

The system of record, the storefront, the directory and the floor all have to agree, and no single team is accountable for the seam between them. That is the work this group takes — named people, not whoever is free that sprint.

A connector library, not a blank editor

Over a hundred pre-built and certified connectors. The hard jobs start from something that already runs against the real API, not from a first integration written under a deadline.

Multi-system, not point-to-point

Three and four systems in one flow, with the master-data direction and the failure behavior settled before the first mapping ships.

Manufacturing and distribution depth

Decades in the industries where a broken interface stops a line or a truck. The edge cases are familiar because we have already been paged for them.

Called in when it is already late

A stalled project, a go-live with a date already promised, a system the last team could not reach. We take the work at that stage, and we take it often.

See Our Success Stories
Initiatives

Most of this work arrives with a project name attached. Start where you are.

Application Development

Keep the system of record. Rebuild the experience around it.

The reason people work around your ERP is rarely the ERP. It is the screen. A receiving clerk should not navigate four tabs to post a pallet, and a field technician should not be handed a desktop form on a phone. So we build the application the job actually needs, on top of the systems you already run, with Magic XPI handling everything behind it.

  • Give us a screenshot, we will give you an app. Your existing screen is a specification, and we treat it that way.
  • Build once, deploy anywhere — one application reaching web, iOS and Android without three codebases to maintain.
  • Existing business logic is reused, not rewritten. The validation rules that took a decade to get right stay where they are.
  • The app writes back through governed integration flows, so there is no shadow database to reconcile later.

Customer outcome

50%

reduction in development costs

Christofle cut development costs in half.

Not by writing less software — by writing it once. One application definition, delivered across the channels their teams actually use, with the enterprise logic behind it left intact.

What we build most often

  • Order entry and quoting for inside sales and dealers
  • Warehouse receiving, picking and cycle counting on handhelds
  • Field service and inspection apps that work without signal
  • Supervisor and approval screens that replace an email chain
  • Customer and partner portals reading live ERP data
AI Development

We do not build AI models. We make AI work.

Our AI engineering is in-house, and it starts from an unglamorous observation: the pilot did not fail because the model was not smart enough. It failed for two reasons, and both of them are data problems wearing an AI costume.

Failure one

It cannot see enough, so it fills the gap.

A model asked about a late order, a margin swing or a line stoppage can only reason over what it was handed. Hand it a PDF export and a document index, and it will still answer — confidently, in a complete sentence, with a number it constructed. That is not the model misbehaving. That is a model doing its job on an inadequate context.

Failure two

The facts it does cite are already wrong.

Standard cost is not actual cost. The routing says the machine runs at a rate it has not held in two years. The ERP has not heard from the floor in days, so on-hand is a guess with a timestamp. Give a model perfect access to that and you get fluent, well-sourced, confidently wrong answers — which is strictly worse than no answer, because now someone acts on it.

The governed answer

Magic MCP sits between the model and your systems.

Magic MCP is a model-agnostic governance layer. It defines which systems a model may reach, which records inside them it may see, which actions it is allowed to take, and what gets logged about every one of those decisions. Because it is model-agnostic, the model you choose this year is a configuration, not an architecture commitment. And because Magic XPI is already connected to the ERP, the CRM, the WMS, the historian and the floor, the context the model reads is live and current rather than exported and stale.

  • Your cloud with controls — governance applied inside your own tenant.
  • Private VPC — no model traffic crossing a shared boundary.
  • Air-gapped on-prem — for environments where nothing leaves the site.

What we actually do

  • Fix the inputs first

    Real production data, real actual cost, current inventory. The unglamorous part is the part that decides whether any of it works.

  • Govern the access

    Scoped permissions, audit trails and human approval on anything that writes.

  • Stay model-agnostic

    We integrate the models you have chosen. We do not sell you ours, because we do not have one.

  • Ship something narrow

    One question, answered correctly and repeatably, beats a platform nobody trusts.

Managed IT Services

Go-live is the start of the obligation. Not the end of it.

Integration that nobody maintains becomes the next generation of brittle scripts. Managed IT services and lifecycle support are how what we build stays working — through ERP upgrades, platform releases, staff turnover and the enhancement requests that arrive the week after launch. It is the same team, under the same P&L, which is why there is no handoff to argue about.

Run and monitor

The integration plane, the applications on top of it and the servers underneath are monitored, alerted and patched on a schedule somebody owns. When a flow fails at 3 a.m., the alert goes to an engineer who already knows the environment.

Lifecycle and version currency

ERP upgrades, platform releases, OS and database end-of-life, certificate rotation. The work that never makes a roadmap and always makes an incident. We plan it and we execute it.

Enhancement after go-live

Every real deployment generates a queue of small changes in its first year. We staff for that queue instead of treating each item as a new statement of work.

Helpdesk and end-user support

Day-to-day IT support for the people using what we built, with escalation into the same engineering team rather than a ticket thrown over a wall.

Proof

Outcomes in production. Numbers the customer will stand behind.

Part of the combined Magic group: 15,000+ professionals, 6,000+ customers, 50+ countries, in business since 1983. What follows is closer to home.

Michelman

60+ hours saved weekly

Specialty chemicals, with ROI in under 10 months. Hours that were going into manual handoffs between systems, returned to the people doing the work.

Kinetico

45 minutes saved per order

Standardized integration between Salesforce and JD Edwards, so an order stopped being a retyping exercise between two systems of record.

Remtron

650% growth in online orders

From 2 orders a day to 15, by connecting the commerce channel to the back office instead of staffing the gap between them.

VB Cosmetics

53% more Black Friday orders

Peak-season order volume up 53%, with zero IT issues through the weekend that breaks most integration estates.

West Coast Reduction

200 vehicles, 150 drivers

Live since November 2022, with 20+ enhancements delivered since — the lifecycle support model working exactly as intended.

Confidential

18 plants in 18 weeks

A repeatable deployment pattern rolled plant by plant, one week per plant, instead of one bespoke project per site.

Certifications and validations

The audit answers, already prepared.

SOC 2 Type II (Feb 2026)ISO 9001ISO 27001GDPROracle Platinum PartnerOracle Validated Integrator — JD Edwards

Integration touches customer data, financial data and production data at the same time, which means security review is not a formality. We bring the evidence to the first call.

One P&L, Three Divisions

Integration is one of three divisions. None of them subcontract the others.

Magic Engineering, Magic Consulting and Magic Integration are divisions of Magic Software Americas, under one P&L. We are the only company in the Americas with OT engineering, enterprise integration, bespoke application development, AI development and operational strategy in-house, with no layer handed to a subcontractor. That is why nobody gets to blame the boundary.

Bring us the two systems that will not talk. We will show you the other nine.

Tell us what is on either end and what has to be true when the data arrives. You will get an engineer on the call, not a qualification script.