MCP for Manufacturing: Connecting AI to ERP and Production Data

What is MCP?

In November 2024, Anthropic released the Model Context Protocol (MCP), an open-source standard that lets AI models, copilots, and assistants connect to external systems to access local files on a desktop or databases.

By connecting tools like ChatGPT and Claude to other systems, these AI tools can access key information and do things like suggest optimizations within a specific workflow or even autonomously execute tasks to fix issues they uncover. Because of how MCP works, it operates similarly to a USB-C port and is a standardized way to connect AI and tools together, so it can pull data from anywhere, like:

• Databases
• Search engines
• Code repositories
• Web services
• File systems

Because MCP enables a virtually endless tech stack for manufacturing, manufacturers have increasingly used it to incorporate AI into workflows like procurement and production.

Why MCP Matters

As we mentioned, MCP matters because it created a universal way for AI to integrate with other software, whereas before each connection had to be custom-built and maintained for each tool and data source.

And to explain how this works, we’re going to have to get nerdy.

The formula for these connections is M x N:

• M = the number of AI applications
• N = the number of tools or data sources being connected to

So, if you have 4 AI applications and 5 tools, there will be 20 connections to build and maintain:

4 M x 5 N = 20 connections

As you can see, the more AIs and tools a company uses, the more connections it needs, and each one had to be a bespoke engineering build to get working, which ultimately led to redundant development efforts and fragmented solutions, since every time there was a change to an API or a new AI model released, the connections would break those builds.

Once a change broke the build, AI models started trying to achieve their goals based on out-of-date information because they couldn’t pull fresh data, so they gave generic answers until someone did a tedious workaround to fix the broken connection.

Fortunately, with the release of MCP, the M x N problem is solved because the communication layer, a client-server architecture over JSON-RPC 2.0, is standardized… if you’d like to know more, we’re sure our engineers can answer more confidently than our writers.

Regardless of our marketing team’s limited technical background, they do understand that, with MCP, each AI host is implemented on the client side once, and the same for each tool or data source.

So, instead of making 20 connections, you make one per AI application and system, bringing those connections down to 9.

MCP solves this M × N problem by standardizing the communication layer through a client-server architecture over JSON-RPC 2.0. Each AI host implements an MCP client once, and each tool or data source implements an MCP server once. The work becomes M + N, so the same 4 AI applications and 5 tools need only 4 + 5 = 9 connections instead of 20. Each new application or tool adds one connection instead of multiplying the total.

Now Explaining Why MCP Matters in English

Yeah, it’s super cool what MCP can enable due to it standardizing communication… but what does that mean for us who are only now getting used to using formulas in spreadsheets?

MCPs are great because your AI assistants can connect to and access your:

• Files
• Calendars
• Code
• APIs

Before, you might have pasted files (shhh... we won’t tell anyone) into an AI tool or tried to find a solution to a problem, but your AI was either stuck with the context you fed it or the latest training data it had access to.

But an MCP connection means you don’t have to explain everything to get answers, because it already has access to your systems and can pull information in real time.

Let’s look at an example.

Imagine you’re on Jira, looking for a ticket. Even though all your colleagues would admit no one really knows how to use any Atlassian product, you’re too scared to ask how to find tickets assigned to you since you’re already 9 months into a job. With an AI connected via an MCP, you can ask it to find all the tickets assigned to you and even draft status updates based on their progress (though they probably won’t be ready, since it’s the first time you’re seeing the tickets).

The use cases don’t need to be job-threatening, but they can be, for example, using an MCP connection to draft updates, with a Google Drive MCP server supplying notes and a Calendar server booking events.

So, that’s everything you need to know about the world of the Model Context Protocol. Now it’s time to get back to what MCP for manufacturing is all about.

What MCP Means for Manufacturers

Ask a random person in the street what manufacturing means, and they’ll likely describe something around assembly lines or automotive manufacturing.

But everyone knows that manufacturing environments never look the same. A huge plant using continuous manufacturing flows might use a custom-built, in-house manufacturing ERP software, while a small shop floor manufacturer might use spreadsheets and an MRP solution. And it’s not uncommon at all for a manufacturer to use all those solutions and more.

For those manufacturers with a tech stack like a Jenga tower ready to collapse, it means a whole bunch of systems that aren’t sharing or operating on the same data.

If you introduce an AI model into the mix, your assistant is going to struggle to achieve any goals since it won’t be able to make sense of the data structure or which tools it should even be pulling data from.

MCP for manufacturing will help you expose data gaps from all your different tools by analyzing:

• MES
• ERP software
• Quality tables
• Machine telemetry
• Unified Namespace (UNS) data
• Maintenance records

As it does this, the AI using MCP for manufacturing connection can attach metadata and definitions to information it reads, so it can start understanding what different things are and what they mean. So, when it reads information showing a temperature spike or a bottleneck, or when it accesses a work order, it will know what it’s reading and what it all means for the business.

With this context, you can ask why a production line slowed down yesterday, log a quality defect in a single sentence, or request the OEE of a workstation over a certain period.

If you’re worried about MCP for manufacturing giving your AI this much access to your business, you can set the safety boundaries.

For example, you can define what to exclude from the tool with commands and restrict what information your AI can see. This means that your MCP for manufacturing can be set up in such a way that the last 7 days of OEE data and error logs are retrievable, but sensitive information remains outside of the AI’s scope.

However, this only works if the data recorded and maintained at the source is accurate. So, as much as you can manage your production with multiple tools, it’s best practice not to, since MCP for manufacturing is only as good as the data it pulls from, and fragmented, siloed, or outdated product records just guarantee that you get bad results from your AI assistant, just really quickly.

MCP Use Cases on the Shop Floor and in the Office

Manufacturing teams already have plenty of data and need help making sense of it.

Once systems are connected through MCP ERP integration, AI can assist and take action in both production and administration.

Shop Floor Use Cases

Operational data on the floor usually sits inside MES, ERP, and SCADA historian systems.

A production planner who would otherwise click through nested screens can ask an agent which machine is the bottleneck today. The same agent can show the late orders for Work Center 2 or report the average cycle time for the last 100 units on Line 4.

For predictive maintenance, middleware adapters convert raw SCADA and PLC sensor tags into MCP-compliant formats. An agent can query that live data and cross-reference heat spikes with historical digital twins to diagnose anomalies before a machine breaks down.

For quality, an operator can log an issue hands-free through an MCP server for manufacturing built on a platform like Sutton. The agent then checks past quality logs and advises whether the defect requires a line stoppage.

Office Use Cases

In the back office, an agent connected to ERP, CRM, shipping software, and vendor email can reconcile purchase orders, PDFs, and inventory levels automatically.

Finance teams can link an MCP for manufacturing server to an AI interface. Non-technical analysts can then pull real-time data or generate data tables with plain-language prompts, without writing complex SQL scripts.

Why MCP Needs a Clean, Connected System of Record

MCP sits on top of core databases, ERP systems, and CRM platforms as a translation and discovery layer.

Those systems keep ownership of their schemas, business logic, and access controls, and MCP cannot reconcile messy data or repair broken permissions.

APIs and databases expect strict, deterministic inputs, while AI models reason probabilistically. A clean system of record lets the backend accept an agent’s query or field change without validation errors or corrupted data.

Documentation needs the same care as the data layer.

An agent that retrieves answers through an MCP server can only return what the source contains, grounding its responses in a live source instead of training data and reducing hallucination. Gaps in the documentation show up as gaps in the agent’s responses. An undocumented feature cannot be described to users, and deprecated features left in place get recommended. Information duplicated across several places with slight variations surfaces as inconsistent answers, so one authoritative source needs to stay current.

Access to the system of record depends on vendors, who have historically offered APIs to meet customer expectations and to make switching harder.

MCP for manufacturing lowers the engineering cost of building against a system and lets any AI agent read its data without a formal partnership. Incumbent systems of record will therefore likely offer basic read access and standard operations while holding back richer data, such as record change history. Some may limit write access entirely, while newer challengers are likely to be enthusiastic about MCP as a way to differentiate themselves.

And there you have it! Everything you need to know about MCP for manufacturing. Now it’s just a matter of finding a manufacturing MCP integration that can work alongside your preferred AI tools.

MCP and Sutton

Sutton is a business operations platform that helps teams centralize and manage:

• Inventory
• Orders
• Supply chain data

It also supports integrations and AI connections, so tools and AI assistants can read live data and help automate workflows.

MCP for manufacturing lets an AI assistant securely access your Sutton data in real time and work with it in chat. Your AI assistant connects to Sutton through an MCP connector. The connector uses OAuth authentication and a server URL found in Settings under MCP. Once the connection is in place, the assistant can read and interact with that data.

The assistant can query:

• Sales orders
• Purchase orders
• Quotes
• Line items

MCP for manufacturing can check inventory levels, warehouse locations, and lot information, and it can browse items and bills of materials. It can also create or update orders, items, and company records. You can ask it to show open sales orders, check inventory, or create a purchase order without leaving the chat.

Instant data access lets you pull metrics and records without opening tables. The assistant can also generate summaries or daily priorities based on your Sutton data. If you work with the API, it can help write and explain API queries for your account.

Want to see this in action for yourself? Book a free demo, and one of our experts will take you on a personalized tour of Sutton so you can see how it will work in your workflows.