AI assistants on your servers.
Through your PAM.
Your engineers' AI assistants are already on their laptops. The moment one needs to touch a server, there are two answers on the table: say no, or hand over the SSH key. Monopam adds the third: the assistant connects to Monopam, not to the server, and what follows is the privileged session you already govern.
- Claude Code · Cursor · VS Code · Claude Desktop
- Zero credentials in the assistant
- Every session recorded
An extension of the person. Not a new user.
Monopam exposes an MCP server that AI assistants connect to with a single URL. The person signs in with their corporate account and approves the assistant once; from then on the assistant works under that identity, inside the policy, recording, and audit you already run.
One URL. Zero credentials.
Setup is giving the assistant an address. The standard does the rest: it discovers the server, registers itself, and signs the person in through Monosign with OAuth and PKCE. There is no password and no key inside the assistant, ever.
- Works end to end with Claude Code, Cursor, VS Code, and Claude Desktop
- Corporate sign-in with OAuth + PKCE through Monosign
- A consent screen the person approves once
- No SSH keys, passwords, or tokens pasted into assistant config
Acts as the person. Never beyond.
The token carries the person’s identity with the assistant’s identity on top, so every policy filter applies to the person. The assistant cannot do anything its owner could not do themselves.
- 27 tools: inventory, access requests, SSH, database, RDP, VNC, and Kubernetes sessions, secrets
- Credentials come out of the vault and go to the target; the assistant never sees them
- Reading a secret is separately audited as a reveal
- Permissions are the person’s, enforced on every call
Recorded like any privileged session.
Sessions the assistant opens appear in the live list with an AI badge, are recorded like every other privileged session, and land in the audit trail under both names: the person’s and the assistant’s.
- AI badge on live sessions and recordings
- Audit rows read "Claude Code (monopam) on behalf of …"
- Admins see every connected assistant and who approved it
- What you show an auditor is what you already show them
Approval gates and a kill switch.
Connections can require approval, and anything that hits a privileged command rule goes to the approval flow: the approval is bound to that exact command and is spent once. When you want an assistant stopped, you do not wait for a token to expire.
- Connection approval and privileged-command approval
- Command approvals are single-use and command-bound
- Admin suspend stops access within two minutes; we measured 91 seconds
- The person can disconnect their own assistant at any time
Connect. Work. Stay in control.
Connect
The developer adds one server URL to the assistant. The browser opens the corporate sign-in, then the consent screen; the person approves the assistant once.
Work
The assistant opens sessions through Monopam: brokered, recorded, written to the audit trail under both names. Privileged commands wait for approval where your rules say so.
Control
Admins see every connected assistant on one screen. Suspend one and its access stops within two minutes; the person can disconnect theirs from their own account.
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.