Google Agent Designer (Gemini Enterprise)

What you'll get

Google Agent Designer is the no-code/low-code agent-building surface in Gemini Enterprise. Once you register the LUU MCP server in your Agent Registry, every agent (single-step prompt agents, flow-builder multi-step agents, scheduled agents) can call LUU's 27 tools natively — alongside Google Workspace, Drive, BigQuery, and any other MCP-registered tool in your enterprise catalog.

Why this channel matters: Pentagon employees built 103,000 agents in five weeks on Google's Agent Designer (Breaking Defense, 2026-04). That's the scale of distribution that opens up once you're a registered MCP tool.

Prerequisites

• Gemini Enterprise subscription (any tier with Agent Registry access)
• Project IAM role — at minimum roles/aiplatform.agentRegistryAdmin
• An LUU API key — get one at the signup endpoint

Architecture overview

┌──────────────────────────┐
│ Gemini Enterprise        │
│ ┌──────────────────────┐ │
│ │ Agent Designer       │ │  user builds agents here
│ │ (flow builder + UI)  │ │
│ └────────┬─────────────┘ │
│          │ tool calls    │
│ ┌────────▼─────────────┐ │
│ │ Agent Gateway        │ │  enforces permissions / scopes
│ │ (Agent Registry)     │ │
│ └────────┬─────────────┘ │
└──────────┼───────────────┘
           │  HTTPS + Bearer token
           ▼
   https://api.leadershipunderuncertainty.org/api/mcp
   (LUU MCP server — 27 tools)

Step 1 — register the LUU MCP server

In the Cloud Console: Vertex AI → Agent Registry → MCP Servers → Add Server

Or use the equivalent CLI / YAML — see agent-registry-manifest.yaml in the repo.

Click Test connection. You should see HTTP 200 and a tools/list response with 27 entries (the canonical luu_* names).

Click Publish. The server is now visible to any agent in your project.

Step 2 — grant access via Agent Gateway

By default, registered servers are accessible only to the publishing project. To open access across your org:

Agent Gateway → Access Policies → Add Policy with role roles/aiplatform.mcpToolInvoker bound to your org domain.

Step 3 — build your first agent

Option A — single-step prompt agent

You are a strategic-intelligence assistant. When a user asks a
strategic question, use the LUU MCP tools (luu_forecast, luu_brief,
luu_fact_check, luu_note_review, luu_get_snapshot, luu_get_delta) as
appropriate.

For probabilistic questions, prefer luu_forecast.
For memo-quality reviews, use luu_note_review with policyFramework='icd203'.
For evidence packs, use luu_search_evidence + luu_fact_check together.

Always cite your sources. Always include calibrated probability language
when stating a forecast.

Option B — flow-builder multi-step agent

Import the included flow template from flow-templates/memo-critic.json:

Agent Designer → New Agent → Import → upload memo-critic.json

This is the equivalent of the n8n memo-critic template, ported to Agent Designer's flow format: user submits memo → luu_note_review → format ICD 203 scorecard → reply.

Step 4 — schedule recurring agents (optional)

Compliance notes

LUU's outputs are standards-aligned and cryptographically attestable:

• ICD 203 — every luu_note_review output is scored against the U.S. Intelligence Community Analytic Tradecraft Standards
• NATO STANAG — source-credibility scoring uses the official 6×6 reliability/credibility matrix
• schema.org/ClaimReview — claim-verification outputs include JSON-LD blocks compatible with fact-checking pipelines
• Cryptographic attestation — every artifact can be SHA-256 hashed, daily Merkle-sealed, GitHub-anchored, and Ed25519 tenant-signed

For DoD or other regulated environments: the substrate has been empirically validated via a 5-phase ladder with pre-registered acceptance thresholds. Cross-domain instantiations exist for medical, legal, and scientific judgment — the substrate is not strategic-cognition-specific.

Pricing

Free tier: 25 calls/day, 500/month, 6 free-tier tools. No card required. Above the cap, HTTP 402 with payment instructions. For higher-volume / enterprise use: per-call credits via USDC or Stripe; subscription tiers; private overlays + BYOD + SSO/RBAC for regulated buyers.

Troubleshooting

"Test connection" returns 401: Bearer token is wrong or in the wrong Secret Manager version. Verify with curl -H "Authorization: Bearer YOUR_KEY" /api/mcp/health.

tools/list is empty: Free-tier nb_* keys see 6 tools; broader access requires a nb_mcp_* key.

Tool calls fail with 429: Rate limit. Stagger calls, request a higher-tier key, or top up credits.

Agent doesn't auto-suggest LUU tools: Be explicit in the system prompt — list tool names directly. Agent Designer's tool-suggestion is more conservative than chat-LLM tool calling.

Next steps

• See the full install markdown for the operator-action checklist (security review, marketplace listing, etc.)
• Try the n8n install guide for the visual-workflow-automation alternative
• Read the tool catalog (JSON) to see all 27 tools + 42 Tier 1 primitives