IBM i holds decades of mission-critical business data (order history, inventory, financials, customer records) and that data is the fuel that makes agentic AI useful. Agentic AI goes beyond chat and code completion: agents can autonomously plan, reason across multiple steps, call tools, and take actions on behalf of users or other systems. Connecting that capability to IBM i’s Db2 for i unlocks a new class of automation that was previously not possible.
Modern AI agents are not just language models. They combine natural-language reasoning with the ability to query systems, run tools, and act on results, closing the loop between insight and action. For IBM i organizations, this creates concrete opportunities across several domains:
Business Process Automation
Data Analytics
System Operations
Security
Performance Management
Two open-source projects from IBM make agentic AI on IBM i practical today.
IBM i is a uniquely SQL-rich platform. System health, job activity, security events, user authorities, storage, performance metrics: virtually everything about the system is queryable through SQL. This means that if an AI agent can run SQL against IBM i, it can do almost anything: read business data, inspect system state, run CL commands, monitor jobs, and more.
Mapepire is what makes that possible for modern AI workloads. It is the server-side component that gives AI agents, MCP servers, and agent frameworks a clean, secure, and modern way to execute SQL against Db2 for i without the friction of traditional IBM i connectivity.
The result: an AI agent connected through Mapepire has access to the full depth of IBM i, expressed through a single, consistent SQL interface.
Mapepire is a prerequisite for the IBM i MCP Server. Install it first:
# On your IBM i systemyum install mapepire-serveryum install service-commandersc start mapepireFull details: Mapepire System Administrator Guide
The IBM i MCP Server is IBM’s open-source bridge between AI agents and IBM i. It implements the Model Context Protocol (MCP), the emerging open standard for connecting AI agents to tools and data sources, and exposes Db2 for i through YAML-defined SQL tools.
How it works: AI clients connect via MCP → the server executes YAML-defined SQL tools → results stream back through Mapepire.
The MCP server ships with ready-to-use tool collections covering:
The tool collections above are a starting point, not a ceiling. Any SQL query your team already runs and trusts (a custom service table view, a business report query, a favourite performance check) can be packaged as a tool by defining it in a YAML file. Once defined, that tool becomes available to any connected AI agent.
This means the SQL knowledge that already lives in your team (the queries your DBAs, operators, and developers have refined over years) can be surfaced directly to agents, without rewriting anything. If your team can write it in SQL, agents can use it.
# Start the server with your custom toolsnpx -y @ibm/ibmi-mcp-server@latest \ --transport http \ --tools ./tools/my-tools.yamlThere are three main paths for agentic AI on IBM i, suited to different organizational needs and technical maturity levels.
IBM watsonx Orchestrate is IBM’s enterprise-grade platform for building and deploying agents that span multiple systems. It provides a no-code to pro-code environment with built-in governance, observability, and pre-built integrations to 80+ enterprise applications including SAP, Salesforce, ServiceNow, and Workday.
Connecting watsonx Orchestrate to IBM i via the MCP Server gives you enterprise orchestration on top of IBM i data. Agents built in Orchestrate can query Db2 for i, trigger workflows, and participate in multi-agent pipelines alongside agents that touch other enterprise systems, all under a governed, auditable framework.
IBM Bob is IBM’s AI-powered development partner, built as an agentic coding assistant for the full software lifecycle. Bob operates in configurable agentic modes: you describe what you want, and Bob plans and executes across your codebase.
Bob connects to the IBM i MCP Server, meaning it can query your IBM i systems directly as part of its development context, looking up schema, checking system state, or helping modernize RPG with live awareness of what’s in Db2.
For teams who want full control over agent behavior, the IBM i MCP Server exposes a standard MCP interface compatible with any MCP-supporting agent framework. Pre-built examples are available in the agents/ directory of the MCP server repository:
| Framework | Language | Strengths |
|---|---|---|
| Agno | Python | Production-ready agents with built-in observability |
| LangChain | Python | Complex workflows and tool chaining |
| BeeAI | Python/TS | IBM’s open-source multi-agent framework |
| CrewAI | Python | Multi-agent coordination and role assignment |
Additional examples can be found in the community db2i-agents repository.
Best for: Teams building custom agents, experimenting with multiple frameworks, or integrating IBM i into a larger open-source AI stack.
A major trucking and logistics provider was struggling with high call volumes and limited support staff — drivers faced long wait times for roadside assistance, repairs, and parts ordering. Working with IBM Business Partner Real Vision Software, the company built an agentic AI-powered voice assistant running directly on IBM i.
Unlike a traditional chatbot, the assistant operates as an autonomous agent: it understands driver intent, reasons across integrated backend systems, and acts without waiting for a human to intervene. In practice, it can:
By integrating directly with core IBM i business systems, the solution handles secure data access, transaction processing, and real-time decision making at enterprise scale — exactly the kind of workload IBM i was built for.
Results: 60% faster resolution of driver support cases, with 24/7 availability and zero wait time.
1. Install Mapepire
Set up the WebSocket database server on your IBM i system.
Mapepire Install Guide →
2. Deploy the MCP Server
Clone and configure the IBM i MCP Server.
Quickstart →
3. Connect an AI Client
Use Claude, Claude Code, VS Code, or IBM Bob as your MCP client.
Client docs →
4. Explore Tool Collections
Browse the built-in SQL tools for performance, security, jobs, and more.
Tools guide →
For enterprise deployments, start with the watsonx Orchestrate page to understand governance and orchestration options before going to production.