What if you could speak in a chat to check what happens in Interoperability, is there any errors, and even solve some types of issues.

With MCP server, you can connect any of your MCP Client, for instance Claude, to IRIS, and ask to check on Interoperability

What you need is a simple configuration of Claude.

• Install Claude Desktop itself
• install uv, which helps with python packages

Claude settings in Developer section, can help to find a config file

With simple setting, like here, it only requires access to IRIS, nothing needs to be installed in IRIS.

{
  "mcpServers": {
    "iris": {
      "command": "uvx",
      "args": [
        "mcp-server-iris"
      ],
      "env": {
        "IRIS_HOSTNAME": "localhost",
        "IRIS_PORT": "1972",
        "IRIS_NAMESPACE": "USER",
        "IRIS_USERNAME": "_SYSTEM",
        "IRIS_PASSWORD": "SYS"
      }
    }
  }
}

After Claude restart, it should display that some MCP tools are available now, click there will display all the tools available

Claude knowing about the tools available, and by prompt you provide, can select which tool best suite to be used to help with the query.

Guess, it works fine, let's check for errors.

We are able to see the actual response from the tool. And Claude analyzed the output and extracted errors it found. Suggesting why the errors may happen. And even offers a ways to solve them.

Currently the implementation of this MCP Server does not offer any code editing capabilities, we are not going to ask to fix the code. But we can try to recover production.

Restart should help

Here we are, all good now.

Well, since our production is working fine now. Let's have a look at the data we may already collected.

To help with it, Claude will use tool which can execute SQL Query.

Even though, I put class name instead of table name, and it discovered that there is no such table, but it did not give up, and in one session, managed to find out the real name for the table, and fetch some data.

When it tries to get some more understanding in the data it collected, and tried to even guess the column names.

Found the real columns, but struggled with reserved keywords, and did not try to use "FOUND"

And with a final attempt

It decided to keep it simple, but using knowledge about column names, it got it right and gave some summary

At this stage, the MCP seems to be in the very beginning of its journey, but still we have quite a good tool.

If you like this tool, please vote on the current OpenExchange Contest