How-to · MCP Server

One URL for the assistant.
Your PAM around it.

Claude Code, Cursor, VS Code, and Claude Desktop can open privileged sessions through Monopam instead of around it. The assistant never holds a credential: the person signs in, approves it once, and everything it does lands in the same recordings and audit trail as any other privileged session.

  • OAuth + PKCE sign-in
  • Recorded with an AI badge
  • Kill switch in minutes
Walkthrough

From a config line to a governed session.

Five steps. No keys or passwords ever enter the assistant.

  1. 01

    Turn on the MCP server in Monopam

    Enable the MCP server under System. From then on Monopam exposes a single endpoint that MCP clients discover on their own, and every assistant that connects shows up as an AI client you can see and manage.

    Tip — The admin view lives at System > MCP Server > AI clients; the same assistants also appear in Monosign under AI Agents.
  2. 02

    Give the assistant the URL

    Add one server address to the assistant’s MCP configuration. The protocol does the rest: the client discovers who guards the resource, registers itself, and starts the sign-in. The same URL works for Claude Code, Cursor, VS Code, and Claude Desktop.

  3. 03

    Sign in and approve the assistant, once

    The browser opens your corporate sign-in through Monosign with OAuth and PKCE, then a consent screen: this assistant will act for you. After the single approval, the assistant holds a token that carries the person’s identity with the assistant’s identity on top.

    Tip — The person can revoke that consent at any time from their Monosign account: Security > Connected AI clients > Disconnect.
  4. 04

    Work through brokered sessions

    The assistant now uses Monopam tools: inventory, access requests, SSH, database, RDP, VNC, and Kubernetes sessions, and secrets. Sessions appear in the live list with an AI badge and are recorded; audit rows name both identities. A command that matches a privileged-command rule waits for approval.

    Tip — Command approvals are single-use and bound to the exact command text; reading a secret is separately audited as a reveal.
  5. 05

    Verify the kill switch

    Suspend the assistant from the AI Agents screen and watch its next call get refused. Access stops within two minutes of a suspend or a user disconnect, without waiting for the token to expire; in our measurement it took 91 seconds.

Ready for the full picture?

The complete walkthrough — with every screenshot, every flag, and version-specific notes — lives in our help center.

Continue in the full docs

Ready to start managing
identities the right way?

Spin up a fully-loaded trial tenant in under five minutes. No credit card. No sales gate.