Overview
Manage your LoginWith organizations from an AI agent.
LoginWith runs a Model Context Protocol server over the console’s management API. Connect an agent — Claude Code, or any MCP client — and it can list your tenants, create OAuth clients, edit login pages, register webhooks and read the audit log, as you, with your permissions.
It is the admin surface only: nothing in it acts as one of your app’s end users.
Hosted server
https://mcp.loginwith.dev/mcpStreamable HTTP. Every request carries a LoginWith access token as
Authorization: Bearer …; your MCP client gets one for you through OAuth the
first time you connect. See Connect a client and
Authentication.
Tools
29 tools, grouped by what they act on:
| Group | Tools |
|---|---|
| Organizations | list_organizations, get_organization |
| Tenants | list_tenants, create_tenant, update_tenant, delete_tenant |
| Members | list_org_members, list_tenant_members, add_tenant_member, update_tenant_member, remove_tenant_member |
| OAuth clients | list_clients, create_client, update_client, rotate_client_secret, delete_client |
| Login pages | list_pages, get_page, create_page, update_page, delete_page |
| Branding kits | list_kits, create_kit, update_kit, delete_kit |
| Webhooks | list_webhooks, create_webhook, delete_webhook |
| Audit log | query_audit_log |
Every tool except list_organizations takes org — an organization id
(org_…) or slug. Updates follow the API’s rule: the fields you pass are the
fields that change.
Each tool calls one management API route and returns
its JSON response as a text block. A route that answers 204 — every delete —
returns OK (no content). A failure comes back flagged as an error, as
LoginWith API error (<status>): <code>: <detail>; no token is ever echoed.