Authentication
Slaunt uses the standard MCP OAuth flow: an unauthenticated request gets a 401 pointing at discovery metadata, and the client registers itself, runs authorization code + PKCE, and retries with a bearer token. Clients like Claude Code, Codex and Cursor do all of this automatically.
HTTP/2 401
www-authenticate: Bearer realm="central-mcp-ccr", resource_metadata="https://central-mcp-ccr-production.up.railway.app/.well-known/oauth-protected-resource/mcp", scope="mcp:tools", error="invalid_token"
{"error":"unauthorized","error_description":"Authorization bearer token is required for MCP tool access","resource_metadata":"..."}Endpoints
All on https://central-mcp-ccr-production.up.railway.app.
| Protected resource metadata | /.well-known/oauth-protected-resource/mcp |
| Authorization server metadata | /.well-known/oauth-authorization-server |
| Authorize | /authorize (email, Google or GitHub sign-in) |
| Token | /token |
| Dynamic registration | /register |
| Revoke | /revoke |
| Grants | authorization_code, refresh_token |
| PKCE | S256 only |
| Client authentication | none: public clients, no secret |
| Scope | mcp:tools |
Scope
The token determines the workspace and project. Many tools accept optional org_id and project arguments, but they exist only as a cross-check: if they don't match the token, the call fails. Leave them out.
Tokens are per client. Signing in to the web app, or to Claude Code, doesn't authenticate Codex. Use identity_context to confirm which workspace a client is signed in to.