Sachi AI
Get updates
All AI offers

AI in your ERP and Microsoft 365, through connectors with set permissions

Secure connectors (MCP) that bring AI agents into your ERP, Microsoft 365, your CRM, databases and internal APIs, each with set permissions.

The problem

An agent can only use the systems it can reach. What it may do in them decides how safe it is. Many trials give the model one shared account with wide access, so one wrong call can reach any record.

Each connector gives the agent a few tools and checks every call

We build MCP connectors. MCP (Model Context Protocol) is an open standard for connecting AI agents to tools and data. Each connector gives the agent a small set of tools. Each tool checks who is asking and what they may reach. It returns only the fields needed and records the call.

Systems it touches

  • Your ERP
  • Microsoft 365: Outlook, SharePoint, Teams
  • CRM platforms
  • SQL databases and internal APIs

Example

A delivery date changed through two separate tools

A made-up case that shows the mechanism. It describes no client.

  1. A procurement agent needs the open orders for one supplier.
  2. It calls the connector’s read tool for open orders, with the supplier’s code.
  3. The connector checks the agent’s role, queries the ERP with a credential from your secret store and returns five fields per order.
  4. The agent then asks to move one delivery date. That tool changes data, so the request waits on the approval screen.
  5. The procurement lead checks the old and new dates and approves.
  6. The connector makes the change. Both calls go into the audit log with the run ID.

It works in the systems you already have

MCP is an open standard. The same connectors serve agents built on provider models and on open models, and they keep working if you change the model. Connectors run inside your network, next to the systems they reach.

Where your data stays

Permissions limit what it can touch

Where a person approves

  • Actions that change data can need approval, tool by tool
  • Your system owner reviews new tools and wider access before they go live
  • Reading and changing data are separate tools with separate permissions. Most agents only get the read tools.
  • Passwords and keys stay in your secret store. The agent never sees a password or a raw key.
  • Every call is logged with the run it belonged to. An auditor can trace any change back to the request that started it.

Everything we build is handed to your team

Who owns the code, and how handover works
  • MCP connectorsCode in your repository, running inside your network.
  • A permission plan, role by roleA written plan your system owners approve and keep.
  • An audit log of every tool requestStored in your own systems, under your rules on how long data is kept.

How we would start

Discovery, then connectors for one task

We list the systems a first task needs and the level of access for each. Then we build and test those connectors before any agent is given them.

How we work together

Questions

What is MCP, in plain terms?

The Model Context Protocol is an open standard for connecting AI agents to tools and data. Connectors built to it are not tied to one AI model or one supplier.

Can you connect to a system with no API?

Often, yes. We can go through its database, its file exports or a small service we build in front of it. We find the safest route in discovery, and we tell you when a system should not be connected.

Is an MCP connector a security risk?

Less than the usual alternative. A shared account gives a model everything. A connector gives it a few tools. Each tool is checked against the role of whoever is asking, and each call is logged.

Name the first system an agent should reach

Say what an agent would need to read or change in it. We will propose the tools and the permission for each one.

Start a project