06 Jul Custom AI Agents in HR Software: What MintHCM Has Already Built
Key Takeaways
- MintHCM already has a live MCPDocumentation module with 73 entries describing module-level business processes (from candidatures to calendar), which Claude already uses today through MCP.
- The AIPromptTemplates module lets a custom prompt be assigned to any of MintHCM’s 119 modules – an element already built into the system schema.
- The native SuiteCRM workflow engine (AOW_WorkFlow) can already trigger actions on record save – the mechanism needed to start an agent automatically.
- A custom AI agent in HR requires three elements: process documentation, a task definition (prompt), and a trigger. MintHCM has built each to a different degree of readiness – what’s missing is the layer that ties them into one interface.
- Every conversation with MintHCM’s existing AI Agent (built on LangGraph, with a human-in-the-loop mechanism) is already logged in the ev_AiConversations and ev_AiMessages modules.
Introduction
Custom AI agents in HR software need three ingredients: documentation the agent can consult, a prompt defining its task, and a trigger that starts it. MintHCM already has fragments of all three built into the product schema. The MCP layer exposes 73 process-documentation entries that Claude already draws on today. What’s missing isn’t the foundation – it’s the layer that connects these pieces into an interface anyone can configure without code.
What’s the difference between a custom AI agent and the AI Agent chatbot?
MintHCM’s AI Agent is a single, pre-defined chatbot – it runs on the LangGraph framework, holds natural-language conversations, and executes actions through the API after user confirmation (human-in-the-loop). A custom AI agent is a different concept: instead of one universal conversational partner, you define multiple narrow, task-specific agents – each with its own prompt, access to specific MCP tools, and its own trigger. The difference is architectural, not cosmetic.
Which pieces of a custom agent has MintHCM already built?
The three elements needed to build a custom agent already exist in the MintHCM schema, each at a different stage of readiness. The MCPDocumentation module provides process context and already holds 73 active entries.
The AIPromptTemplates module provides the task-definition mechanism – its root_module field lets a prompt be assigned to any of the system’s 119 modules. The AOW_WorkFlow engine provides a trigger based on record save. Access to MCP tools, which Claude already uses in everyday recruitment work, is a piece a custom agent wouldn’t need to build from scratch. The table below shows the detailed status of each element.
| Custom agent element | What it’s for | Status in MintHCM today |
|---|---|---|
| Process documentation | Teaches the agent a module’s business logic | MCPDocumentation – 73 entries, actively used by Claude through MCP |
| Task definition (prompt) | Defines what the agent should do | AIPromptTemplates – module ready, assignable to 119 modules, 0 records |
| Trigger | Starts the agent automatically | AOW_WorkFlow – native workflow engine, 2 defined processes |
| Tool access | Lets the agent act, not just answer | 16+ MCP tools – already used by Claude and the AI Agent |
| Agent-building interface | Ties the above into one no-code configuration | Doesn’t exist yet – the direction ahead |
How does process documentation already teach Claude MintHCM’s logic?
MCPDocumentation works like a set of notes that Claude consults when a module’s logic isn’t obvious from the data structure alone. Entries in the system are grouped by module – from Candidatures and Competencies, through Calendar and Check availability, to Exit Interviews and Reports.
The exact content of individual entries isn’t published. But the range of categories alone shows that the documentation already covers most key recruitment and employee processes, not just a single test case.
What’s still missing to call this an agent-building platform?
Process documentation and the prompt mechanism both exist, but each is at a different stage of adoption – see the table above for details. What’s missing is the piece that would tie documentation, prompt, trigger, and MCP tool access together into a single “agent” record.
That record should be configurable without writing code. It’s this step – not the absence of the building blocks themselves – that separates today’s state from the full realization of the AI-enabled open source HCM vision MintHCM has been communicating since 2025.
Why does this incremental architecture matter for IT and HR decision-makers?
Open source code means an IT department can verify each of these modules itself, without waiting for vendor confirmation. In closed HR systems, the ability to define agents depends entirely on what the vendor chooses to expose.
A closed-system customer has no visibility into whether process documentation or a prompt mechanism even exist, let alone the ability to extend them. In MintHCM, these elements are part of the same codebase the organization self-hosts and can modify if a regulation or business process changes.
Frequently Asked Questions
What is a custom AI agent in HR software?
It’s a task-specific AI agent defined through configuration – its own prompt, access to specific tools, and a trigger – as opposed to one universal chatbot handling every request.
Is MintHCM’s AI Agent the same as a custom agent?
No. AI Agent is a single, ready-made chatbot built on LangGraph with a human-in-the-loop mechanism. A custom agent is the concept of multiple narrow, configurable task-specific agents – a different architecture.
What is MCPDocumentation and how does Claude use it?
It’s a MintHCM module with 73 entries describing the business logic of individual modules. Claude consults it through MCP when the data structure alone doesn’t explain a process – for example, unusual relationships between records.
Can I already define my own AI agent in MintHCM today?
Not fully as a no-code feature. The building blocks exist partially – the prompt module and the trigger mechanism are built, but they aren’t yet connected into a single configurable agent record.
Does MintHCM’s approach protect against vendor lock-in when building AI agents?
Yes, structurally. All the pieces – documentation, prompts, triggers, MCP tools – are part of open source code the organization self-hosts and modifies itself, rather than a service bought exclusively from one vendor.
What is AOW_WorkFlow and how could it trigger an agent?
It’s the native workflow engine from SuiteCRM, the framework MintHCM is built on. It can run an action after a record is saved – exactly the mechanism an agent would need to start automatically, for example after a meeting is created.
Why does this matter for AI Act transparency requirements?
The AI Act (Regulation EU 2024/1689) requires organizations using high-risk systems to be able to show how the decision-making logic works. Open process-documentation and prompt code makes that easier to demonstrate than in a closed system.
Sources
Regulation (EU) 2024/1689. Official Journal of the European Union, 12 August 2024.