How it works
Three steps. The agent never sees your user's password.
The AI agent asks to connect
A user adds your MCP server to Claude, ChatGPT or another AI agent.
The user signs in and grants permissions
Authgear shows your login page and asks the user what the agent may do.
Your server lets the agent in
The agent gets a token that only works on your server. Your server checks it on every request.
Why teams use Authgear for MCP
Works with the AI tools your users already use
- Claude
- Claude Code
- ChatGPT
- Codex
- Cursor
- VS Code
Frequently asked questions
Do I need CIMD or DCR?
Turn on both. CIMD is the method the MCP spec now recommends: the agent's ID is a URL pointing to a document the agent hosts. With DCR, the agent registers itself, and some clients still rely on it. Authgear supports both, so every agent can connect.
Which AI agents work with it?
Any client that follows the MCP authorization spec, including Claude, Claude Code, ChatGPT, Codex, Cursor and VS Code.
Is it on the free plan? Is it in the open-source version?
Yes to both. MCP support is included on every Authgear plan, including free, and in the open-source build you can self-host.
How are agents billed?
They aren't billed separately. An agent acting for a user doesn't count as an extra monthly active user.
Can I use it with my existing users?
Yes. If your users already sign in with Authgear, agents use the same accounts, login page and roles. Register your MCP server as an API resource and turn on AI agent access.