---
title: "Permissions and security"
description: "How the MCP connection uses your own Vigilis permissions, what the three access scopes mean, and how to review or revoke a connection."
---

> Documentation Index
> Fetch the complete documentation index at: https://support.vigilis.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Permissions and security

An MCP connection is an extension of your own account, not a service account with broad rights. Everything below follows from that.

## The connection runs as you

When you connect an assistant, you sign in with your own Vigilis credentials and the connection inherits your role. If you cannot open a record in the Vigilis interface, the assistant cannot reach it either. If your organization requires SSO or MFA, you pass that check at connection time like any other sign-in.

This also means the connection is personal. It is not shared with your teammates, and it does not survive your account being removed.

## The three access scopes

During sign-in, your client requests one or more scopes. Your client shows you which ones before you approve.

| Scope | What it allows |
| --- | --- |
| `mcp:read` | Read your organization's records: invoices, inventory, contracts, circuits, tickets, and reports. |
| `mcp:write` | Create and change records, such as opening a ticket or submitting a change request. |
| `mcp:admin` | Administrative reads, including deleted records held back from ordinary lists for merge, restore, and audit work. |

Scopes set the ceiling. Your role sets the floor. A connection granted `mcp:write` still cannot write to anything your role does not allow, so a member with read-only access in Vigilis stays read-only through MCP. For how roles work, see [Roles and permissions](/navigation/roles-and-permissions/) and [Members and roles](/administration/members-and-roles/).

If you only intend to ask questions, approve a read-only connection where your client offers the choice.

An administrator can also cap a connection at read-only from the Vigilis side. When that is set, every call from that assistant is treated as a read no matter which scopes it holds, and writes are refused. It is the right setting for an assistant you want asking questions but never changing records.

Two further administrative controls are worth knowing about. A connection can be pinned to operate as one named person, in which case its reach is computed from that person's live role and memberships and can never exceed what they already have. And access can be withdrawn mid-session: a revoked connection's next request is refused, not just its next sign-in.

Separately, nothing a write tool does takes effect on the first call. Creates, updates, deletes, and bandwidth changes all return a preview and wait for an explicit confirmation. See [What your assistant can do](/mcp/what-your-assistant-can-do/).

## What leaves Vigilis

This is the part worth being deliberate about. When your assistant retrieves a record, that record is sent to the AI provider running your assistant so it can answer you. That provider is your choice and your contract, not ours. Vigilis does not select it, does not receive a copy of your conversation, and has no control over how the provider handles what you send.

Practically:

- Treat an MCP query the same way you would treat exporting the same data and pasting it into that assistant.
- If your organization restricts which AI tools may handle carrier or financial data, that policy applies here.
- Vigilis logs the access itself, so administrators can see that data was retrieved.

## Reviewing and revoking access

Vigilis keeps its own list of what you have connected, and you do not need an administrator to manage it.

Open your profile and find **Connected Applications**. Each assistant you have authorized appears once, showing the vendor, the scopes it was granted, when you granted them, when it was last used, and whether it is still active. **Revoke** ends it immediately.

Revoking here is the reliable way to cut access. Removing the connector inside your AI client stops that client from using the connection but leaves the authorization standing in Vigilis, so do both if you are removing an assistant for good.

One connector often registers itself with Vigilis many times over, so a single Claude connection can sit behind a dozen internal registrations. The Connected Applications list collapses those into one entry per assistant, and one Revoke drops all of them together.

Changing your password, disabling your account, or removing you from the organization all end the connection as well. If you see an entry you do not recognize, revoke it and contact support at [support@vigilis.io](mailto:support@vigilis.io).

## Common questions

**What are the mcp:read, mcp:write, and mcp:admin scopes?**

They are the three levels of access a connection can request. Read lets the assistant look at your data, write lets it create and change records, and admin covers organization-level settings. Your Vigilis role still limits what any of them can actually reach.

**Can the AI assistant see data from other customers?**

No. The connection is scoped to your organization and runs as your user account, exactly like your browser session.

**How do I revoke an assistant's access?**

Open your profile, find Connected Applications, and click Revoke on the assistant. That ends it immediately without needing an administrator. Removing the connector in your AI client as well is good practice.

**Does connecting an assistant send our invoices to an AI company?**

The data your assistant retrieves is processed by whichever AI provider you use, under your agreement with them. Vigilis does not choose or control that provider. Treat an MCP connection the same way you would treat pasting a report into that assistant.

Source: https://support.vigilis.io/mcp/permissions-and-security/index.mdx