Decided once
Which assistants may reach which systems, for whom, and with how much access: settled centrally, not person by person.
Trustaige · Platform · AI access · Preview
Say yes to AI assistants. Keep every key in your hands.
Your people are already connecting AI assistants to company systems. Trustaige lets you approve each assistant once, give it no more than the person using it holds, bind its tokens so a copied one is useless, see every request it makes, and end it in one click.
AI at work
When your people bring AI to work, it has to follow the same rules they do.
Nothing quietly overreaching
No token pasted into a chat log that still works. No assistant reaching further than the person using it. No access nobody remembers granting.
A yes you can defend
Your teams get the assistants they want, and you have the record that shows exactly how they’re used.
On every AI connection
Four guarantees, enforced by Trustaige
rather than by the assistant.
01
Approved once, for everyone
Link your organization’s own AI client once. Everyone signed in to it through Trustaige reaches the tools you allow, without each person approving it again, and only an assistant that can prove who it is can be linked.
Under the hoodMCP enterprise-managed authorization, built on the IETF’s Identity Assertion Authorization Grant, a working draft.
02
Only what the person can reach
A connection carries only what the role of the person behind it already permits, and that is checked again on every request. Change the role and the assistant’s reach changes with it.
Under the hoodScopes capped by the person’s live role and read again on every call. Access tokens last an hour.
03
A copied token is useless
Turn on key binding and every grant is tied to a key only the assistant holds, and can be used once. A token lifted from a log or a screenshot can’t be replayed by anyone else.
Under the hoodSender-constrained tokens with DPoP, RFC 9449.
04
One click to end it
Revoke a connection, unlink an AI client, or suspend the person who approved it, and its tokens stop working immediately. Nobody has to chase a vendor.
Under the hoodEvery call is checked against live state, so ending access needs no expiry window.
Built in · Preview
Ask your directory a question, in plain language.
Trustaige runs its own read-only MCP server for your directory, devices and audit trail. Your security team asks an assistant and gets answers from live data instead of exporting reports.
- Read-only, always. No tool on it writes. An assistant can answer questions; it can never change your directory.
- Works with the assistants your team already uses. Claude, Claude Code, VS Code, or any client that speaks the Model Context Protocol.
- About five minutes to set up. The client discovers everything from one address and sends the person to Trustaige to approve.
On the record
Every AI request, accounted for.
See which AI clients reached which systems, on whose behalf, and what was refused and why. Approving, revoking, linking and unlinking land in the same hash-chained audit log as everything else, and stream to the monitoring tools your team already runs.
When the board asks how AI is being used on company data, the answer is a report, not an estimate.
See audit & governanceBeyond your own tools · Preview
The same control over other vendors’ AI connections.
When an app you use runs its own MCP server and supports enterprise-managed authorization, Trustaige governs that too. Your people’s assistants reach it under your rules, not the vendor’s defaults.
- 1Paste the server’s address
Trustaige reads the server’s published details and fills in the rest.
- 2Keep only the access you grant
Scopes the vendor adds later stay off until you switch them on.
- 3Your policies still apply
A person must be assigned to the app and pass your access policies before any assistant acts for them there.
For your engineers
Open specifications, all the way down.
Nothing here is proprietary plumbing. Every guarantee above is an open specification your engineers can read, test and hold us to.
curl https://id.trustaige.com/.well-known/oauth-protected-resource/mcp- MCP authorization
- Model Context Protocol specification, including enterprise-managed authorization
- Identity Assertion Authorization Grant
- IETF OAuth working group Internet-Draft (ID-JAG)
- OAuth 2.0 Token Exchange
- RFC 8693
- JWT bearer authorization grant
- RFC 7523
- DPoP
- RFC 9449
- Protected resource metadata
- RFC 9728
- Resource indicators
- RFC 8707
- Client ID metadata documents
- IETF OAuth Internet-Draft
Start a conversation
If your auth layer is on the agenda, so are we.
We'll walk through a working deployment, map it to your stack, and tell you honestly where Trustaige fits — and where it doesn't.