Article
· Mar 7 2m read

Building AI Agents with interSystems IRIS - What do we mean?

Hi, colleagues!

As you can see the new topic of the programming contest - AI Agents. 

The topic was over-hyped recently on the Internet and has different meanings. You might get curious about what we mean by AI agents in regard to the InterSystems programming contest.

Indeed, we believe that AI agents can change the landscape of IT solutions in almost all spheres significantly boosting its efficiency and often changing the way of management, consumption and operation.

In general, AI agents are about the enablement of different business processes into automation so that automation includes decision-making based on any of the GenAI engines available on the market, e.g., OpenAI, Claude, or Grok which could be used in conjunction with IRIS Vector Search related to RAG solutions. 

Usually, AI agent building assumes leveraging any of the scenario automation engines, such as Zapier, Make, N8N, Pydentic or/and InterSystems IRIS Interoperability.

What AI agents can do? Well, we can ask ChatGPT about it now, but the most usual practices are:

  • - checking, filtering, and answering emails,
  • - looking for events of a kind and booking tickets,
  • - answering support tickets,
  • - monitoring database/services and cleaning up or providing service operations.

So in this contest, we'll consider as an AI Agentic a solution that leverages:

- any LLM engine for making decisions (e.g. OpenAI, Antropic, Grok),

- any automation/interoperability engine (e.g. Zapier, Make, N8N, IRIS Interoperability),

- either IRIS database, IRIS Interoperability, or IRIS Vector Search usage is a mandatory condition.

Good luck, and I am looking forward to introducing more helpful AI agents of any kind powered by InterSystems IRIS!

Discussion (4)3
Log in or sign up to continue

Short version: MCP provides a standard means to integrate GenAI with the data it needs to be helpful in the user's context (resources) and the things it can do (tools). Plus a few other things.

Especially since OpenAI has joined the bandwagon with MCP, it seems like it's here to stay.

Bottom line, I think there's a fantastic role for IRIS Interoperability to play with MCP. I'm hacking around with it a little bit personally - though as an InterSystems employee I won't be submitting my hacks to the contest, of course.

There are three major use cases I see:

IRIS as an MCP client - orchestrating activities across MCP servers with traceability and in a centralized way (which could be of value in an enterprise setting - one chatbot backed by IRIS interop that has access to all the right resources with proper access controls). This is where I've been playing around. In case anyone else is too: I've had more luck with SSE than stdio due to some Embedded Python oddities I haven't had time to fully explore, but this is probably better architecturally anyway: put the MCP server in its own container that you connect to rather than worrying about having IRIS call out to Python.

IRIS as an MCP server - do things in IRIS and get access to your data in IRIS with MCP (I've caught wind of one awesome project along these lines already...); the scope of this could also include making it easy to enable your own IRIS-based application, whatever it is, as an MCP server.

IRIS as an MCP proxy - why not both? To support use cases like Claude Desktop where you want to work against local files and such but also don't want each person in the company setting up and updating their own set of 20 MCP servers, re-expose all of the appropriate tools/resources/etc. (with proper access controls and perhaps governance over LLM use by resource/tool due to data sensitivity, etc.) as a single MCP server everyone can connect to.