How to connect an AI agent to RentProg

An AI agent works with company data on your behalf over the MCP protocol: it sees the same branches, records and fields as you do, and no more. The “MCP for AI agents” service must be enabled in the company's plan.

Two ways to connect

Method	For which agents	Is a key needed
OAuth connector	Claude in the browser and in the desktop app, ChatGPT, Claude Code with the RentProg plugin	No: you sign in to RentProg and choose access on the consent screen
Personal rpa_ key	RentProg CLI, Cursor, Codex and other clients without OAuth	Yes, issued in the profile

Both methods are set up in the profile, in the “AI agent keys” section. It also shows a hint about which connector to add, the “Connected applications” list and the list of keys.

Where to get the connector address

The connector name and address for your company are shown in the profile, in the hint of the “AI agent keys” section. Use exactly those.

How to connect the connector in Claude

1. Open the profile, the “AI agent keys” section, and look up the connector name and address in the hint.
2. In Claude's connector settings, add a custom connector with the name and address from the hint. On Claude team plans, the connector is added by the organization owner.
3. Click connect — the RentProg sign-in opens.
4. On the consent screen, choose access by area and click “Allow”.
5. Turn on RentProg in the conversation you need through Claude's tools menu.

How to connect the connector in ChatGPT

1. Turn on developer mode in ChatGPT's app settings if it is not on yet.
2. Create a new app with the connector name and address from the profile. If ChatGPT asks for an authorization method, choose OAuth.
3. Start the tool search — the RentProg sign-in opens.
4. On the consent screen, choose access by area and click “Allow”.
5. Wait for the tools to load and save the app.

In a ChatGPT workspace, an administrator may need to publish the app for members.

How to connect Claude Code

1. Install the RentProg plugin version 3.0.0 or later:

   /plugin marketplace add RentProg/rentprog-claude-plugin
   /plugin install rentprog@rentprog

2. When the plugin asks, enter the connector address from the profile. The plugin does not need a key.

3. Open /mcp, select the rentprog server and complete the sign-in and consent in the browser.

4. Check that /mcp shows the rentprog server with a list of tools.

Tip

If you had the 2.x plugin with a key, update it. Revoke the key in the profile if it is no longer used anywhere else.

What you choose on the consent screen

Access is set separately for each area: CRM, fleet and bookings, money, employees. The same levels are chosen when issuing a key.

Level	What the agent can do
No access	Does not see the area
Read	Only views data
Write with approval	Prepares a change, which is carried out after your approval in the “Awaiting approval” section
Write with preview	First gets the result of the change, then confirms it itself
Write immediately	Makes the change without preview or confirmation

Levels above what your role allows are not offered. If the employee's permissions are later reduced, the agent's access shrinks along with them. For money we recommend “Write with approval”, for CRM — “Write with preview”. Reconnecting the same app replaces the previous access.

How to review and approve an operation

With “Write with approval” access, the agent prepares a change first. Open “Awaiting approval” in the RentProg menu or from the notification about a new operation.

1. Check the name in “Operation” and the connection in “Key”.
2. Read “What will be done”: fields have clear labels such as “Client”, “Booking”, and “Cash desk”. Follow links to related records to check the source data before deciding.
3. Check amounts, dates, and recipients, then click “Approve” or “Reject” in the row. Approval starts the operation on your behalf.
4. For several operations, select the rows, click “Approve selected” or “Reject selected” (the button shows the number of selected rows in brackets), and confirm. You can process up to 50 operations at a time.
5. Check the status and explanation in each row: rejecting one operation does not cancel the other decisions.

The menu item is available to the “Employee”, “Manager”, “Branch manager”, and “Main Administrator” roles. Each user sees and decides only on operations prepared on their behalf in the current company group. The “Main Administrator” can revoke other users' keys but cannot approve their operations. User and connection permissions are checked again when the operation runs.

Important

You have 24 hours to decide; the deadline appears in “Expires”. After that, the operation becomes “Expired” and is not performed. If the result has changed since preparation, the status becomes “Outdated”. Nothing has been performed — ask the agent to prepare the operation again.

“Done” means successful completion, “Rejected” means the user declined, and “Not performed” means execution failed. Read the explanation in the row. “Accepted” and “In progress” do not confirm completion yet: wait for the final status.

Updating the list without reloading the page

Statuses update automatically, including when a decision is made on another device. Processed rows stay on screen until you refresh the list, so you can check the results.

If new operations arrive while you are selecting rows or reviewing results, “New operations awaiting approval have arrived” appears. Click “Refresh” when you finish reviewing. The list will show pending operations, and the previous selection and results will disappear from this screen. If there is no current work or results, new operations load automatically.

How to recognize an agent's action

The “Via AI agent” robot icon next to a record means it was changed through an AI agent. Hover over it to see the tool and time of the action. These record markers are available to staff. For a CRM message, the tooltip may only say “Sent via AI agent”.

The marker does not name the employee. To see which employee the agent acted on behalf of, compare the time with the activity log, if your role and subscription plan give you access. An outgoing CRM message also shows its author when known to the system.

How to issue a personal key

1. Open the profile, the “AI agent keys” section, and click “Create key”.
2. Enter the key name and access by area.
3. Copy the key right away: after the window is closed, it is not shown again.
4. From the same window, copy the MCP server address and, for agents with a terminal, the ready-made CLI command.

Important

A key is valid for one year, no more than five keys can be active at the same time, and revoking a key cannot be undone. Apps connected via OAuth do not count toward this limit. Keys are not issued to partners and agents.

With a key, the agent connects to the MCP server address from the profile and sends the header Authorization: Bearer rpa_…. For example, in Claude Code without the plugin:

claude mcp add --transport http rentprog address_from_profile --header "Authorization: Bearer rpa_your_key"

How to revoke an agent's access

1. Open the profile, the “AI agent keys” section.
2. Find the app in the “Connected applications” list or the key in the list of keys.
3. Click “Revoke access” next to the app or “Revoke” next to the key, and confirm.
4. If needed, also remove the connector in the Claude or ChatGPT settings.

A revoked app or key loses access on the agent's very next request to RentProg. The main administrator sees the access of all employees in the company group in both lists and can revoke any of it.

Where the data goes

Everything the connection sees, including clients' personal data, the agent passes to the AI service you chose. Your company chooses the service, and it is responsible for making sure such a transfer is permitted by the laws of the country and by its agreements with clients.