IBM i MCP Server
An open-source, IBM-published Model Context Protocol server that lets AI agents and LLMs query Db2 for i data, inspect system objects, and take defined actions on IBM i using natural language.
The IBM i MCP Server is an open-source project published by IBM on GitHub, implementing the Model Context Protocol (MCP), an open standard for connecting AI agents and LLM-powered tools to external data and actions. For an IBM i shop, it means an AI agent can query Db2 for i data, inspect system objects, and take defined actions on IBM i, all through natural language, rather than requiring hand-written SQL for every question.
How it connects to Db2 for i
Mapepire connectivity
Uses Mapepire, a secure, modern Db2 for i connectivity layer, to reach Db2 for i data rather than an older or ad hoc connection method.
YAML-defined tools
SQL tools are defined in YAML, shared with a companion ibmi CLI (@ibm/ibmi-cli), so the same tool definitions can be used from an AI agent or from the command line.
MCP-compatible clients
Works with any MCP-compatible AI client, including Claude, GPT-4-based tools, and local LLMs, rather than being tied to one specific AI vendor.
What it is not
This is not a machine learning or predictive-modeling feature. It does not train or score models, and it is not a substitute for a demand-forecasting or anomaly-detection capability. It is an agent-connectivity and tooling layer: it lets an external AI agent or LLM query and act on Db2 for i data using natural language.
The IBM i MCP Server is in beta, opened October 27, 2025. Confirm current capability and production-readiness against IBM's GitHub repository before scoping a project around it.
Related
IBM Bob (Project Bob)
IBM's generally available AI coding agent for IBM i. Bob 2.0 and the Bob Premium Package for i shipped June 24, 2026, adding a Db2 for i Database Mode, about 30 IBM i tools, and direct work on source where it lives. Reported pricing from $20 to $240 per month.
IBM watsonx.ai
IBM's studio for building, tuning, and deploying foundation models and generative AI applications, and the underlying AI layer other watsonx products are built on.
IBM watsonx Code Assistant for i
IBM's November 2025 preview announcement of an IBM i-specific, RPG-focused AI code assistant. This preview track was superseded by IBM Bob, which reached general availability in March 2026.
IBM watsonx Assistant
IBM's general-purpose conversational AI platform for building chatbots and virtual assistants. Not IBM i-specific, but relevant for shops building conversational front ends over IBM i or Db2 for i data.
Frequently Asked
What is the IBM i MCP Server?
It is an open-source, IBM-published server implementing the Model Context Protocol (MCP), letting AI agents and LLMs such as Claude, GPT-4, and local LLMs query Db2 for i data, inspect system objects, and take defined actions on IBM i using natural language, over a Mapepire connection. It is in beta since October 2025. See our IBM i MCP Server explainer for how it differs from a machine learning feature.
Is the IBM i MCP Server a machine learning feature?
No. The IBM i MCP Server is an AI agent connectivity layer ... it lets an AI agent or LLM query and act on Db2 for i data using natural language over the Model Context Protocol. It does not train or score predictive models, and it is not a substitute for a dedicated machine learning or forecasting capability.
Where do the embeddings live for RAG on IBM i?
Outside Db2 for i, on a second system. Db2 for i has no vector store as of September 2026, so there is nowhere in the database to put them. IBM's stated direction, from a May 2026 strategy session, is OpenSearch running on Power as the hybrid vector, text and structured search layer, with Db2 for i staying the system of record and OpenSearch holding the embedded, indexed copy an agent searches. Note that IBM said on Power, not on IBM i: in practice that means a Linux on Power partition, and IBM has published no supported IBM i build, reference architecture or timeline. watsonx.ai supplies the embedding models and the text embeddings API that create the vectors, and IBM's own documentation is clear that storing them is your job, in a vector database you provide. See our RAG on IBM i explainer for the full loop and the order to build it in.