CODESYS Adopts Model Context Protocol (MCP), Challenging Belief That In-House Models Are Best for Industrial AI Deployment

Industrial automation software supplier CODESYS is bringing Artificial Intelligence (AI) to market by integrating Model Context Protocol (MCP) to give customers their choice of Large Language Model (LLM). Originally developed by Anthropic, the MCP is a tool designed to allow AI platforms into internal activities, but does not feed that data back into the broader data training ecosystem of the model’s parent company. Taking this approach has made CODESYS, a company that was behind in AI, competitive with Programmable Logic Controller (PLC) engineering leaders Siemens, Emerson Electric, and Schneider Electric.

CODESYS’ competitors have been making large investments to bolster their AI offerings: Siemens has committed over €1 billion in AI investment over the next 3 years, as well as a US$10 billion acquisition of Altair Engineering; Emerson acquired AspenTech for US$14.2 billion to provide a more robust Advanced Process Control (APC), DataOps, and industrial AI backbone; and Schneider Electric unveiled an agent for its EcoStruxure Automation Expert at Hannover Messe. While ABB, Siemens, and SUPCON support MCP, these companies have also undertaken large AI investments. Meanwhile, CODESYS has avoided such large Capital Expenditure (CAPEX) and can provide access to the models that customers are already familiar with by leveraging support for MCP from OpenAI (ChatGPT), Google (Gemini), Meta (Llama), and its creator, Anthropic (Claude). 

MCP is capable of autonomously running 20 of the services CODESYS offers, providing agentic operations of control engineering such as PLC code generation and testing. MCP-driven control engineering supports localized versions of public models, preventing company data from escaping into the wider ecosystems. Agentic capabilities are the first step to realizing the next innovation of multi-agent systems. Orchestrating agents across services and third-party agents will enable autonomous operations, done through protocols such as Agent-to-Agent (A2A) and Agent Communication Protocol (ACP).

CODESYS’ adoption of MCP also underscores the shift away from internal AI model development in favor of third-party innovation, unrestricted by the relationship management of co-innovation frameworks. While MCP is a more accessible alternative to co-innovation, it hinges on the success of external LLM producers, limiting quality management compared to an in-house solution.

The big question arising from the CODESYS and MCP deployment is: how necessary is an internal AI model? While Siemens has pushed to build its Industrial Foundation Model for AI, Emerson Electric offers DeltaV AI and AspenTech, and Schneider Electric’s recent agent launch uses Microsoft Foundry. CODESYS, leveraging MCP, may be beneficial to companies of its size.

Suppliers that have been slow to develop an AI strategy must act quickly. More competitors are bringing new AI solutions to market,  customers are beginning to scale out of pilot projects, and brand affinity for AI solutions is building.

It is possible for companies skeptical of AI to take a limited/no AI stance, representing an option for the skeptics in manufacturing, which, with the slow-to-adapt nature of customers, may be a tenable position in the near term. The ultimate risk here is that being skeptical leads to being completely uncompetitive, resulting in atrophy of an anti-AI customer base, while also missing advancements in AI technology.