Can the agent see data I am not allowed to see?

No. Tool access is filtered by your permissions, user by user and organization by organization. The connector bypasses nothing: it applies the same rules as the interface.

Do we have to hand you a model key?

No, and that is an important difference from the AI built into the platform. Here your assistant calls its own model with its own credentials. OverView only exposes tools.

Can the agent change our data?

It can create metrics and dashboards, as you would in the interface, and always within your permissions. It does not touch your source tools, which remain the system of record.

Do we need to open a port or install anything?

Neither. Your assistant makes an outbound HTTPS call to a stateless endpoint, and there is no binary to deploy.

How do we cut access off?

From your organization settings. The connector is removed with no side effect on the platform or on your data sources.

Claude Desktop, ChatGPT, Le Chat and Cursor are tested. Any client that speaks the Model Context Protocol can connect; the tested list is not exhaustive.

MCP CONNECTOR

Your AI assistant knows nothing about your infrastructure. Connect it to your inventory.

You add OverView as a connector in your AI assistant, with your own credentials. It then queries your inventory and your metrics, and the answer lands in the conversation, not in one more console.

Endpointhttps://platform.oversoc.com/mcp

Start with your own data What the agent can see ↓

One connector, however many tools sit behind it
The agent only reaches what your account allows
Nothing to install: an HTTP endpoint

WHAT CHANGES

1connector to configure, not one per tool

4clients tested, and any MCP client works

0software to install, no model key to hand over

WHY IT MATTERS

The best tool is the one your team already has open

One connector, not one per tool

The alternative is one connector per source and per person: a separate setup, a separate credential set and a separate access review for each one. And most of your tools expose no MCP server at all.

Your tools finally talk to each other

A machine goes by a different name in the directory, the EDR and the CMDB. The unified asset carries every one of its identifiers, so your assistant knows which names to look for in your other connectors, instead of missing it or searching blind.

Fewer accounts to hand out

Answering one question used to require access to the EDR, the CMDB and the scanner. Source credentials stay inside the OverView connectors: your people no longer need an account in every tool, only their OverView scope.

WHAT THE AGENT CAN DO

The same tools as the platform, exposed as they are

The Model Context Protocol (MCP) is an open standard for connecting an AI application to external systems. OverView speaks it: the servers behind the gateway are merged into a single tool list, and your assistant sees one connector, not a stack of plumbing.

Query the inventory

Machines, users, software, vulnerabilities, network services: search and filter run against the reconciled inventory, not one isolated console.

Run your metrics

Execute an existing KPI from your library and read the result, with the detail behind the number.

Cross-check with your other tools

In the same conversation, line your inventory up against your tickets, your documentation or your other systems. That is the cross-check the platform alone cannot do.

Go back in time

A snapshot of the inventory is taken every day and kept for three months, twelve months or the life of the contract, depending on your plan. Your assistant queries a past date, or compares two dates to see what moved. An EDR only ever shows today’s state.

WHAT THE AGENT CAN SEE

Your scope, and nothing beyond it

Plugging an assistant into your infrastructure is exactly the kind of decision a CISO wants to be able to examine. Here is what it rests on.

Filtered by your permissions

Tool access is controlled user by user and organization by organization. An agent querying OverView on your behalf sees your scope, and nothing else.

No model key to hand over

Your assistant calls its own model. OverView exposes tools; it does not sit between you and your AI provider.

Rate limited

Calls per minute are capped, so an agent stuck in a loop never turns into a production incident.

Stateless

The endpoint keeps no session. Every call is authenticated and authorized on its own.

Tied to a named user

Every call goes through the user’s own OverView account, with their permissions and rate limit. No shared service account: on OverSOC’s side, every request is logged with the user who made it.

Revocable

You withdraw access from your organization settings. Cutting the connector breaks nothing in the platform.

Connected in three steps

No binary, no agent, no port to open. Your assistant calls an HTTP endpoint.

01In OverView, generate your organization’s access credentials.
02In your assistant, add a remote connector and paste the endpoint.
03Ask your first question. The OverView tools show up in the conversation.

Talk to an engineer →

Tested clients

Claude Desktop
ChatGPT
Le Chat
Cursor

And any client able to speak MCP — the list above is not exhaustive.

FREQUENTLY ASKED

What your CISO will ask

The Model Context Protocol is an open standard that lets an AI application connect to external systems. Its documentation compares it to a USB-C port for AI: a standardized way to plug an assistant into data and tools. OverView exposes its own tools through that standard.

THE OTHER ROUTE

Would you rather stay inside OverView?

OverView Workbench does the same work from the platform, on your own model key: you describe the metric, it builds it.

Discover Workbench →

Ready to take back control?

Plug in your sources, see for yourself in 14 days. Free access, no commitment, 45-minute setup.

Start for free