Bring Your Own AI Agent: Exploring the IBM Maximo MCP Server

As we continue our exploration of the IBM Maximo Application Suite (MAS), our next stop is not an application you open. It is a door. The IBM Maximo MCP Server arrived with MAS 9.2 in June 2026.

If your AI strategy so far is a ChatGPT tab open next to Maximo, this one is for you. The MCP server is how AI that talks about maintenance becomes AI that can look up the pump, read its history and raise the work order, through Maximo and under Maximo's rules.

First, What is an MCP Server?

MCP stands for Model Context Protocol, an open standard for connecting AI assistants to the systems that hold the real data. Anthropic published it in November 2024 and handed it to the Linux Foundation a year later. Today ChatGPT, Claude, Microsoft Copilot, Gemini, VS Code and Cursor all speak it.

Think of a restaurant. The system (the MCP server) publishes a menu of things it is willing to do; the AI (the MCP client) reads the menu and orders, and the kitchen does the cooking. Each item on the menu is a tool: a name, a plain-English description, an input form, and flags that say whether it only reads or changes things.

For the Maximo veterans: the Integration Framework taught Maximo to talk to ERPs. The MCP server teaches it to talk to AI agents.

What is the Maximo MCP Server?

It is IBM's own MCP server for Maximo Manage, shipped with MAS 9.2. It runs as a pod on OpenShift beside Manage, and it is available whenever Manage is available. You do not have to set up the Maximo Assistant to build tools for it.

• Bring your own agent - the agentic Maximo Assistant uses it, and so can IBM Bob, Claude, Copilot or an agent your own team built. Connecting is the server's route URL plus one header: a Maximo API key (apikey) for integrations, or a JSON web token (x-access-token) when the agent acts for different people.
• Maximo security, not a new one - every tool is a Maximo REST API call, and the Assistant works within the same security, permissions and approval rules as Manage. Two users asking the same question can get different answers, depending on their security groups.
• Manage owns the menu - Manage is the single source of truth for the tool list, and the server asks it for changes every few minutes (every ten by default). Restart the MCP pod if you cannot wait. The listmcptools API on Manage shows every published tool.

What's in the Toolbox

Most MCP tools call Maximo Manage APIs and you decide which APIs are exposed as MCP tools.

Five types:

• AI configuration - six AI Service models can become tools: data retrieval (natural language to a Maximo query), similarity, two classification models, condition insights and IBM documentation search. An activated configuration is deployed as a tool automatically. This is the only type that needs the AI Service.
• Object structure - a single operation cut out of an object structure. Maximo actions: change status, create follow-up work order, enter meter reading, move asset, change location, report downtime and create service request. You add your own in the Object Structures application, under More Actions, Add/Modify Object Structure Tools.
• Automation script - the big one. Write an active script, add one variable per input, then use More Actions, Add/Modify MCP Tool. The variables become the input form, and a variable named href hands the script the record itself. The team that writes your scripts can now build your AI tools.
• Workflow - run a non-interactive workflow against a record, set up from Workflow Designer using a similar Select Action option.
• API route - IBM-provided system tools with no setup screen, such as the schema retrieval tool that tells an agent how Maximo's objects relate.

Every tool carries a description, and the agent uses it to decide when to call the tool.

Then there are the flags, and here Maximo's meaning differs from the MCP standard. IBM says to tick Read only when a tool does not edit data or its edits are reversible, such as a status change. Destructive hint is for permanent edits, such as deleting a record, and it is the flag that makes the assistant ask for approval first.

Seamless Integration with Other MAS Applications

Manage: The source of truth and the rule book. Every change an agent makes is a normal Manage transaction, executed under user’s own identity. No side door.

Maximo Assistant: Agentic in 9.2: a supervisory agent, the ALM agent, plans your request, calls tools on the MCP server and writes the answer.

Health, Monitor & Predict: The closest link today is the condition insight tool, which IBM describes as analyzing asset health, maintenance history and failure patterns. Every tool IBM documents for 9.2 runs through Manage; there are no Monitor or Predict tools yet.

Reliability Strategies: Publish the related object structures to see which PMs are covering the reliability strategy failure modes.

Considerations to Bring Your Agent

Pointing a public agent such as Copilot, Claude or ChatGPT straight at the server is tempting. Your people already use it, there is no new screen to roll out, and every model improvement arrives for free. The catch is that the agent, not Maximo, decides much of how the work gets done:

• Approval lives in the client. Maximo’s flags are only hints. Whether the agent asks first depends on how the client is set up and, on its instructions (prompts), which can get long and vague.
• Identity is on you. With one API key, every change carries the same name. Acting as each person takes identity work on your side.
• Your data travels. Query results go to the agent's model provider. Check where that is and what it keeps.
• Records can talk back. A work log is text the model reads, and text written to look like an instruction can steer an agent that holds write tools.

The industry is moving toward a safer layer between AI agents and business systems: agents that sign in as the person using them, and gateways that put every MCP server behind one front door for sign-in and logging.

That governed layer is what we built the interloc.ai Agentic EAM Assistant to be: a team of specialist agents operating over a Maximo MCP Server, for teams who want an agent but cannot hand Maximo to an unsupervised one.

• Everyone is themselves. It opens from inside Maximo, so there is no second sign-in. Requests run with the person’s own Maximo access, and their security groups decide which agents they can use.
• The software decides when to ask. Rules in the software, not the AI model, decide which actions run straight through, and which stop for a person’s approval.
• Guardrails in the software, not the prompt. Read and write limits and masking of personal data are enforced in code, so a clever prompt cannot talk its way past them.
• A trail you can audit. Every prompt is logged, and each answer shows which agents and tools produced it.

Where to Start

• Start with tools that only read. Turn on data retrieval and schema retrieval, watch what users actually ask for a month, then add the writing tools one at a time.
• One identity per agent. Give an outside agent its own Maximo user, API key and security group, never the integration superuser. The change history will carry the agent's identity, and it will only see what you gave it.
• Make the agent ask. Only the destructive flag tells an assistant to ask first. Configure whatever client you connect to confirm before any tool that changes data or put a governed agent in front of the server that does.
• Keep your own trace. “Manage” keeps records of who changed what. For an outside agent, logging the conversation behind each call is your job.

Final Thoughts

"AI in Maximo" meant whatever IBM built in. With 9.2, Maximo becomes something any AI can call, while Maximo keeps the keys. The plan is to ship more tools in the coming patches and more of them opened to the Assistant, so the menu will grow. Pick one question your planners ask every week, expose the two or three tools that answer it, connect the agent your people already use, and let it take the first crack.

With that I will sign off for now - stay tuned for more insights into the innovative features and functionalities of the Maximo Application Suite!