Beta Of MCP Server Opens Up IBM i For Agentic AI

October 27, 2025 Alex Woodie

When Anthropic quietly released Model Context Protocol (MCP) in November 2024, it didn’t garner a whole lot attention. But as the agentic AI wave grew in early 2025, the protocol suddenly was hailed as the new standard for integrating AI applications, tools, and data sources. A good indication of just how far MCP has come is the fact that IBM released an early version of the MCP server for IBM i this month.

Project Bob got the big headlines at the IBM TechXchange conference, which took place the second week of October in Orlando, Florida. And for good reason, as Bob is the big new vehicle for Big Blue’s co-pilot ambitions, including for IBM i. While we don’t have a timeline for delivery of a Bob-based coding co-pilot, the early version is apparently so advanced that it led IBM to kill plans for Watson Code Assist for IBM i, which was in public preview this summer and had been slated to ship before we all sang Auld Lang Syne.

But at TechXchange, IBM quietly released another product that has big implications for IBM i’s AI future: an MCP server. IBM’s open source development team, led by business architect Jesse Gorzinski, have developed an early version of a program that allows the IBM i server to communicate with the world via MCP.

Anthropic designed MCP as a lightweight architecture to enable developers to build secure, two-way connections between the MCP server (data sources) and LLM-powered applications (MCP client). Prior to MCP, developers often had to build custom connectors for each data source or tool, which resulted in a mess of different techniques. MCP was developed to serve as a USB-C port for AI applications, according to Anthropic.

According to the documentation on the ibmi-mcp-server Github page, the open source software is designed to serve as a “production-grade” MCP server that allows AI agents to access IBM i database operations through YAML-configured SQL tools. The program is “purpose-built” for enterprise environments, with “comprehensive authentication, observability, and security features.”

The IBM i MCP server, which is distributed under an Apache 2.0 license, uses the new Mapepire client for Db2 for i that IBM debuted a year ago, and supports local (or stdio, short for standard input/output) and remote (or HTTP streamable) connections, according to the documentation. OpenTelemetery support is baked in for tracing, metrics, and performance monitoring. The MCP server for IBM i, which is 72 percent Typescript and 21 percent Python, is type safe and has standard error handling built-in, the documentation states. (Don’t you wonder what the other 7 percent is?)

The architecture of the product was developed to ensure a “clean separation” between business logic and transport handling, using the “logic throws, handler catches” pattern. When an MCP client request is received, it invokes a handler, which validates the input and creates a request and a logic function call. If the logic successfully executives, it’s transported back to the requestor. If not, it generates an error. That’s a gross overcomplication; for a more detailed description, check the documentation.

“This architecture has been battle-tested in enterprise IBM i environments and follows industry best practices for production MCP server deployment,” the documentation on GitHub states. “The modular design ensures that each component can be developed, tested, and maintained independently while working together as a cohesive system.”

The MCP server is designed to help IBM i developers create AI agents and agentic workflows that touch IBM i servers. IBM says developers are to use Agno, an open source Python-based framework, to develop the agentic workflow. Some of the early clients that the MCP Server for IBM i has been used with include Claude Code, Claude Desktop, VSCode, Cursor, Windsurf, Roo Code, LM Studio, OpenCode, Gemini CLI, Cline, and Python Clients, including via Agno and through the MCP’s project official SDK.

Image courtesy IBM.

The MCP server has been in the works for a while. IBM i chief architect Steve Will indicated the software was in the platform’s future in an interview with IT Jungle at the COMMON POWERUp conference earlier this year in Anaheim.

At that time, Watson Code Assist for IBM i was still moving towards general availability. That would have been an MCP client that could have worked with an MCP server on IBM i, among other code assistants. Now that the next-gen AI-based IDE Bob is on the agenda, Bob will feature an MCP client that can talk to the MCP server on IBM i to do things.

While IBM doesn’t intend Bob to be used for systems management tasks on IBM i, Will said he wouldn’t be surprised if IBM i customers used it that way. There will likely be other AI assistants developed that will allow IBM i professionals to use natural language to get information about what’s going on IBM i, such as what PTFs have been applied, without going through Navigator or Access Client Solutions, Will says.

MCP will be the conduit for those other AI assistants or tools to access IBM i, says Yan Zhuo, an IBM i product manager leader with IBM.

“That’s another way to talk about the MCP,” she said. “It’s for automating some of the tasks that are typically done by system administrators, but you can do so using plain language instead of writing SQL codes.”

The MCP server for IBM i project currently has 11 stars on GitHub, six contributors, and two forks. No timeline was given for general availability, but an IBM document states that “our intent is to deliver 500 tools in 2026.” You can read more about the MCP server for IBM i here.

This Issue Sponsored By

Table of Contents

Content archive

Recent Posts

Subscribe

Pages

Search