Knowledge Gateway

Permission-aware access to organizational knowledge for people and AI clients. Sign in with your Google Workspace account; every answer is filtered to what you are cleared to see.

AuthGoogle Workspace (OIDC) — deny-by-default
Health/healthz
Diagnostics/validate (sign in, check your access, run a test query)
REST API/api/v1 (session, packages)
Connect your AI client (MCP)

Claude Code / Claude Desktop / Cursor — any streamable-HTTP MCP client. Add to .mcp.json in your repo root (or your client's MCP config):

{
  "mcpServers": {
    "knowledge-gateway": {
      "type": "http",
      "url": "https://gateway.myhybridlab.com/mcp",
      "oauth": { "callbackPort": 33418 }
    }
  }
}

On first use your client opens a browser window — sign in with your Google Workspace account. No tokens to copy. The callbackPort is required: the sign-in service allows exactly this loopback callback.

Signing in is not access. Any Google account can complete sign-in — that is deliberate, so partners and external collaborators can be granted access without a separate identity. What you may actually read is decided afterwards by your group membership, and an account with no granted groups sees nothing at all.

Authentication is broad, authorization is not — anyone may sign in; only granted groups confer access.
Need-to-know — results are filtered by your clearance; anything you are not entitled to is simply absent.
Audited — every retrieval is recorded in a tamper-evident audit log.
No credentials here — this page never issues tokens; sign-in always happens via Google in your own client.