Architecture and security

Give AI Network Access Without Giving It Everything

The Scrutinizer MCP Server requires authenticated access, respects existing user permissions, and disables write access by default. Users start with read-only access, with write capabilities enabled only when needed.

Remote AI clients authenticate using administrator-issued Scrutinizer API tokens. Access follows each user’s existing roles and permissions, and activity is audited by user so administrators can see who performed each action. The MCP server introduces no separate permission model.

Learn How

How a Request Moves

01 Your AI client sends the request

The user asks a network question in an MCP-compatible client.

02 MCP selects an available Scrutinizer tool

The AI client selects one of 60 available tools based on the task.

03 Scrutinizer checks authenticated access

The remote client presents its API token. Available data and actions stay scoped to that user.

04 Scrutinizer runs against customer network data

Scrutinizer queries the flow data, reports, alarms, and retained history already available in the deployment.

05 The client presents the result

Scrutinizer returns structured output. The AI client explains it in its own interface.

Security Controls That Matter

ControlWhat it means
AuthenticationRemote clients use an API token created by a Scrutinizer administrator.
Permission scopeThe token carries permissions and roles like a user account. The assistant stays inside them.
Read-only defaultAI-assisted work is read-only out of the box.
Write controlOne administrator setting enables all write-class tools. The setting is off by default.
Activity attributionRemote-assistant activity is logged against the API token.
Client responsibilityThe connecting client controls its own model, prompt handling, retrieved context, and any approval experience.

How is my network data used?

The Scrutinizer MCP Server queries network data locally within your environment. All underlying network data stays in your deployment, while the connected AI client uses the query results to generate answers.

Deployment options without vendor lock in

Teams can use an MCP-compatible external AI client, and Plixer AI supports managed or customer-controlled model. Customer-controlled OpenAI-compatible endpoints can support self-hosted, sovereign-cloud, or air-gapped deployments.

Your AI, your model, on your network data. No new data plane and no new permission model. The AI you already run can answer from the network you already built.

Review the connection requirements

Frequently Asked Questions

Is the MCP server part of Plixer AI?

Yes. Plixer AI uses the same MCP server and tools available to external AI clients. This allows external clients to access the same Scrutinizer reporting and investigation capabilities used by Plixer AI.

Can a remote AI client act as a Scrutinizer user?

Yes. An administrator creates an API token assigned to a specific Scrutinizer user. The AI client operates within that user’s existing permissions, and activity is audited under their identity.

Can the AI change Scrutinizer settings?

Not by default. Write access is disabled unless explicitly enabled by an administrator.

Is MCP available in every current edition?

MCP is included with the current Scrutinizer, Plixer One, and Plixer One Unlimited editions. Availability depends on the licensed edition, not just the installed software version.