Agents
Connect AI assistants such as Claude to BeeSensible, so they can look up your decisions on AI tools, request a tool for an employee and register agents in the AI register.
With Agents you connect AI assistants such as Claude to BeeSensible, through the BeeSensible MCP server. MCP (Model Context Protocol) is an open standard that lets an AI assistant use the data and functions of another service. Once connected, the assistant can check what your organisation decided about an AI tool, request a tool on behalf of an employee and register an agent in the AI register. An agent is an AI application that works for the organisation with its own instructions, its own access to systems or the power to act by itself, such as a custom GPT, a Copilot Studio agent or an automation. Because such agents do not use a browser, the extension does not see them; by registering, they still end up in the register.
What a connected assistant can do
An assistant connected to BeeSensible has four functions available:
| Function in the assistant | What it does |
|---|---|
| May I use this AI tool? | Looks up the decision about a tool, with the condition, your text under "Why not?" and the contact person for AI tools. |
| Ask for an AI tool | Requests a tool on behalf of the employee, in the same way as the form in the extension. The request carries the employee's name and email address. |
| Register this agent | Registers an agent in the AI register, or reports it in again at the start of a session. |
| My registered agents | Lists the agents the connected employee registered, with your decisions. |
When it connects, the assistant also receives a short instruction: check a tool before using it or sending data to it, register at the start of every session when acting as an agent, and only request a tool when the employee asks for it. No separate instruction or skill is needed in the assistant. A request made through the assistant appears in the Inbox on the Requests tab, just like a request from the extension; see Tool requests and contact person.
What the assistant hears about a tool follows your choices under Policy > Notices, so an assistant hears the same as an employee in the browser. Nobody is held back from a tool without a decision. If you have switched on the brief notice for No decision yet, the assistant tells the employee that there is no decision yet and takes care with personal data and confidential information. For a tool that is not allowed, the assistant follows your choice of whether an employee can continue anyway.
What an agent states when it registers
An agent registers with a name, the tool it runs on and what it does. It can also answer the following questions; anything it leaves open is shown in the dashboard as "Not stated", which is not the same as no.
| Field | Possible answers |
|---|---|
| What the agent may do | Only read or process information, Suggest actions, Carry out actions itself |
| Approval by a person | Before every action, Afterwards, None |
| Works with personal data | Yes, No |
| Confidentiality | Public, Internal, Confidential |
| Systems and data | Free text, for example which systems the agent can reach |
| Supports decisions about people | Yes, No |
The department comes from the BeeSensible account of the employee who registers the agent; the agent is not asked for it. That employee is the agent's owner and is shown by name under Registered by. Only admins see that name.
A registration is a statement by the agent or its maker, not something BeeSensible observed. That is why a registration does not count in any figures under Usage, Compliance or Accountability; only Insights > Agents shows the registrations, and it says that they were stated.
Where agents appear in the dashboard
An agent sits in the AI register under the tool it runs on: an agent in Copilot Studio belongs to Microsoft Copilot Studio, an agent in Claude to Claude. You find agents in three places:
- In the Inbox, on the Agents tab, are the agents waiting for a decision. The guide for connecting BeeSensible is there too.
- On a tool's page there is an Agents tab with the agents that run on that tool. The Agents | Insights switch moves between the list and the insights about those agents.
- Under Insights > Agents is an analysis of all registrations: which agents need attention, what agents may do set against the confidentiality of what they use, which systems they name, and an overview per platform, per department and per month. There are no decision buttons here; every part links to the Inbox or the tool's page.
You decide on an agent with Allow, Do not allow or Decide later, and Remove from the register removes it. An agent can never be allowed further than the tool it runs on, so you can only allow it once the tool itself is allowed or allowed with limits. The other way round is possible: on an allowed tool, an agent can be not allowed. A condition belongs to the tool, which is why agents have no "allowed with limits".
Connecting BeeSensible to Claude
The guide is in the dashboard, in the Inbox on the Agents tab, in the card Connect BeeSensible to Claude. It also shows the Connection address and the Client ID, each with a button to copy it. The dashboard only lists Claude, because that is the platform the connection has been proven on.
Add a custom connector
In Claude's admin settings, add a custom connector under Connectors.
Enter the address and client ID
Enter the connection address from the dashboard. Under the advanced settings, enter the client ID; no client secret is needed.
Let employees connect
Employees then see BeeSensible among their connectors and choose Connect once. Each employee signs in with their own BeeSensible account, so the register knows who registered an agent. A shared account therefore does not work.
Set the connector's permissions
Set the connector's permissions in Claude. Set Read-only tools and Register this agent to Always allow, and Ask for an AI tool to Needs approval (the hand), because a request goes to the admin in the employee's name. In Auto mode, Claude may go ahead without asking even with Needs approval.
Claude sometimes renames its screens. If a step no longer matches what you see in Claude, let BeeSensible know at [email protected].
Insights > Agents says Agents is not switched on
The Agents module is not on for your organisation. Switch it on under Settings > Modules. If the module is not listed there, the beta has not been switched on yet; request it at [email protected]. Also check that the AI register module is on, because Agents does not work without the register.
An agent cannot be allowed
An agent can only be allowed once the tool it runs on is allowed or allowed with limits. Decide on the tool first, for example through the link to the tool next to the decision on the agent.