Skip to content
Artrilogic

MCP servers

Let automation use your systems, safely and unchanged.

Most of the business runs on systems you cannot rewrite. We put a controlled connection in front of each one, so automated processes can read from it and act on it with the permissions you grant and nothing more. The system itself stays as it is.

Request a Fit callSee agentic AI
AI agents and AI tools, using commercial or private models, reach your existing systems (such as SAP, Dynamics, a mainframe or in-house applications) only through MCP servers we build. Each server checks identity, applies permissions scoped to the step, enforces rate limits and writes an audit record. MCP defines how AI tools call a system; these controls are part of the servers we design, not features of the protocol.

AI tools and agents

  • AI agentsMulti-step automated work
  • AI toolsCommercial or private models

MCP servers we build

A thin adapter in front of each system

  • Identity
  • Scoped permissions
  • Rate limits
  • Audit

Systems you already run, unchanged

  • SAP
  • Dynamics
  • Mainframe
  • In-house apps

MCP defines how AI tools call a system. Identity, permissions, rate limits and audit are controls we design into each server, not features of the protocol.

01What you are trying to do

You want AI to work with the systems that run the business, and those systems were never built for it.

Copying SAP screens into a chat window does not scale, and neither does hand-coding a custom adapter for every new AI tool.

02What gets in the way

Without a shared connection layer, every new AI use case rebuilds the same plumbing.

Each version is built slightly differently, so audit gets harder and permissions drift apart. That shows the first time an auditor or regulator asks who could do what.

03How we help

One controlled connection per system, used within the permissions you grant.

That connection can serve the next use case too. Each new one still needs its permissions set, its integration tested and its capacity checked, and those determine what it costs to add.

How a call reaches your system

One controlled path between AI tools and your systems.

The controlled connection is a Model Context Protocol (MCP) server: a small service in front of the system that AI tools and agents call instead of the system itself. We design it so each call is identified, checked against the permissions for its step, held to rate limits and logged before anything reaches the system behind it.

A call outside its permissions does not get through, and the attempt is recorded. The system itself does not change.

Currently shipping

MCP server estate for an Australian SaaS platform. Controlled access for AI tools, with identity, rate limits and audit logging at the integration layer.

How a call from an AI tool reaches your system through an MCP server: the AI tool or agent makes the call; the server checks the caller's identity; it checks the permissions scoped to that step; it applies rate limits and writes an audit record; only then does the call reach your system, which stays unchanged. A call outside its permissions is refused and recorded.
  1. 01AI tool or agentCommercial or private model
  2. 02IdentityWhich automation, on whose authority
  3. 03Scoped permissionsOnly what this step needs
  4. 04Rate limits and auditCapped, and every call recorded
  5. 05Your systemUnchanged behind the connection

Outside its permissions: refused and recorded

Why we build on MCP

What you get from doing it once, properly.

  • Your existing systems stay as they are

    The MCP server is a thin, governed adapter in front of the system. Your SAP, Dynamics, mainframe or in-house app keeps doing what it does, and the investment already made in it keeps paying back.

  • Every call checked and recorded

    We design it so every automated call passes through identity, permissions scoped to its step, rate limiting and audit logging before it touches the underlying system, so you can see which automation did what, and on whose authority.

  • One standard interface for any AI provider

    MCP is the emerging standard for letting AI agents call tools and read data in a structured way. Building on it makes it easier to change model vendor without rebuilding your connections.

  • Deployed where you decide

    The MCP servers run where you decide, in your own cloud network (VPC) or your data centre. Together with the permissions you grant, that governs which data can leave. Who has operational access and which jurisdiction applies are agreed as separate questions.

Got a system you want automation to reach?

A free 30-minute call with a senior engineer. Bring one problem. We tell you whether it is worth a longer conversation, and if not, who you should talk to instead.

Request a Fit callSee it on WSO2