An industrial products manufacturer · 2026 – present
An internal agent and MCP platform for a manufacturer
A reusable suite of internal agents, skills and MCP servers that put the company's own systems — including its ERP — behind a language model.
Context
A language model is only useful inside a business once it can reach that business's data. The gap is almost never the model. It is that the ERP, the document stores and the internal services have no interface an agent can safely call.
The problem
Give internal teams AI tooling that operates on real systems rather than on text somebody pasted in, without handing a model open-ended access to production.
What I did
- —Built MCP servers that expose specific, bounded operations from internal systems — including IFS ERP — as named tools, rather than opening a general-purpose database connection and hoping.
- —Packaged the work as an internal marketplace of agents, skills and plugins, so a new use case starts by composing existing components instead of commissioning another integration.
- —Kept each server's surface deliberately narrow: named operations with typed inputs, so what the agent is able to do stays a review decision rather than an emergent one.
Outcome
Internal teams get AI tooling that reads and acts on real business systems, and each new use case composes servers that already exist instead of starting another integration from zero.