Sindan Solutions
FRFree quote

Model Context Protocol

MCP servers that let AI agents work inside your business software

We build the MCP server for your business software: the tools an agent can call, the permissions it gets, and the log of what it changed. Ours already runs in production, on Dido.

Discuss your project
An AI agent taking in invoices, emails and forms and turning them into approved tasks.

What we deliver

The work behind an MCP server agents can actually use

Exposing an API is not enough. An agent needs tools named the way the business talks, errors that tell it what to do next, and limits it cannot cross.

  • Splitting the software into tools an agent understands: one module, one prefix, one help tool per module.
  • Connecting the agent to a real user account, with that user's permissions and no more.
  • Hiding write tools from anyone who only holds read access.
  • Showing a preview of every sensitive write before the server saves it.
  • Accepting the names people use for products, customers and warehouses, so the agent never needs an internal ID.
  • Returning errors that name the right next call, so the agent corrects itself.
  • Keeping a log: who, through which connection, which tool, and whether the call worked.
  • Testing the server with real AI agents on real tasks before go-live.

Our own server

Dido, our MCP server in production

Dido is our distribution management software. Its MCP server covers 28 modules: catalogue, stock, quotes, orders, invoices, deliveries, returns, purchasing, cash and bank, credit control, vans and rounds, batches and expiry, production, equipment hire, reports and user administration.

An agent works in Dido as a named person on the team. It sees only the tools that person's role allows, in the warehouse where they work. Dido records every change an assistant makes, and the team can read back who changed what.

See Dido
A digital vault surrounded by a key, a backup and an activity log.
The person's permissionsThe agent inherits the user's role. A tool it may not use does not even appear.
Preview before writingThe agent sees an invoice or a bulk import line by line before Dido saves it.
A readable logDido records every call that changes something: person, connection, tool, result.
One warehouse at a timeThe agent works in one establishment and has to switch explicitly to reach another.

Your case

Three starting points

You sell software

Your customers want to plug in their AI agent. We build your product's MCP server around your roles and your permission model.

You run an internal tool

An ERP, a CRM or a home-grown database. We open it to an agent on a narrow scope, read-only first.

See AI agents

You do not have the software yet

We build the system and its MCP server together, as we did for Dido.

See custom software

Common questions

What people ask us

01What is an MCP server?

The Model Context Protocol is an open standard that lets an AI assistant (Claude, ChatGPT and others) call the functions of a piece of software. The MCP server is the part of the software that exposes those functions, with their rules.

02Do we have to rewrite our software?

No. The MCP server sits on top of your existing logic or API. It adds a layer that talks to agents, checks permissions and keeps the log.

03How do you stop an agent from doing damage?

It acts as a real user, with that user's permissions. The server hides forbidden tools, sends sensitive writes through a preview and logs everything the agent changes. We often start read-only.

04Do you work with clients outside Algeria?

Yes, remotely. The team is based in Algeria on UTC+1, the same hours as Paris and one hour ahead of London. We work in English and French. You pay in euros or US dollars, by Wise, PayPal or bank transfer. Details are on the contact page.

05How much does it cost?

We do not publish prices. The amount depends on how many modules to expose, the permission model and the state of the existing API. After scoping you receive a quote split by phase.

Guides

Further reading

Artificial intelligence · 8 min

Nine rules we follow in the MCP server of Dido, our ERP

How we built the MCP server of Dido, a distribution ERP with 28 modules, so AI agents can sell, invoice and move stock safely: help tools, permission-scoped tools, preview then confirm, business names, errors that name the next call.

Read the guide

All guides

Which software do you want to open to agents?

Tell us what it does and what an agent should do in it. We reply with a first split into tools and the permissions to plan for.

Get in touch