MCP servers and integrations

AI connected to your systems, with control.

MCP (Model Context Protocol) is the open standard for giving AI models access to data and actions from a system. We build custom MCP servers for ERP, CRM and legacy systems, with specific actions, permissions and logging.

The problem

Auto-generated MCP servers expose too much data.

Without proper authentication or audit trails. For company data, that's risk, not a shortcut.

What we deliver

Secure integrations, not generic access.

Custom MCP server

Well-defined business actions (query order, balance, status), not generic database access.

ERP and legacy integration

TOTVS Protheus and other ERPs, databases and internal systems, via API or intermediary service.

Security

Authentication, per-user permissions, server-side validation and human approval for write actions.

Deployment in your infrastructure

The server runs inside the company environment and works with ChatGPT, Claude and custom agents.

How we work

Four steps, from idea to operating system.

1

Assessment

We assess feasibility, data, risks and architecture.

2

Proof of concept

We validate the solution with real data.

3

Production

We build, integrate and put it into operation.

4

Evolution

We monitor, tune and add capabilities.

Who it's for

For companies that want AI using real ERP data.

✓ Para empresas que

  • Companies that want AI assistants and agents using real ERP data.
  • Without switching systems.

Frequently asked questions

Common questions about MCP servers

What is an MCP server?

It's a service that publishes, in a standard format, the actions and data that a system offers to AI models. Assistants like ChatGPT and Claude and custom agents can then use those actions the same way.

Does it work with TOTVS Protheus?

Yes. We integrate via the ERP's REST APIs or through an intermediary service when the API doesn't cover what's needed. Writing directly to the ERP database is out of the question.

Can the server run in our infrastructure?

Yes, that's the standard for sensitive data.

Can AI change data in the ERP?

Only if that action is published, with permission and, at first, with human approval. We recommend starting with queries only.

How to protect an MCP server?

Specific actions instead of generic ones, authentication, role-based permissions, server-side validation, rate limiting and logging of every call.

Read also

Start with the Assessment.

Before you invest, find out whether the idea is feasible, what it costs and what the path is.

Assess my AI project