┌─────────────────────────────┐
                         │    chat.conveyorclaims.ai    │
                         │   (Cloud Run Domain Mapping) │
                         └──────────────┬──────────────┘
                                        │ HTTPS
                                        ▼
┌───────────────────────────────────────────────────────────────────────┐
│                    Cloud Run: hf-chat-ui                              │
│                    ghcr.io/huggingface/chat-ui-db                     │
│                    Port 3000, 2Gi RAM, 2 CPU                         │
│                    us-central1, VPC: conveyor-connector               │
│                                                                       │
│  ┌─────────────┐  ┌──────────────┐  ┌─────────────┐  ┌───────────┐  │
│  │  SvelteKit  │  │  MCP Client  │  │  Multi-LLM  │  │  MongoDB  │  │
│  │  Frontend   │  │  (Tool Call) │  │  Provider   │  │  Sidecar  │  │
│  └──────┬──────┘  └──────┬───────┘  └──────┬──────┘  └───────────┘  │
│         │                │                  │                         │
└─────────┼────────────────┼──────────────────┼─────────────────────────┘
          │                │                  │
          │                │          ┌───────┼───────────────┐
          │                │          │       │               │
          │                ▼          ▼       ▼               ▼
          │       ┌──────────────┐  ┌──────┐ ┌────────┐ ┌─────────┐
          │       │ MCP Bridge   │  │OpenAI│ │ Google │ │Anthropic│
          │       │ (Cloud Run)  │  │ API  │ │Gemini  │ │ Claude  │
          │       │              │  │      │ │ API    │ │ API     │
          │       │ Routes to:   │  │gpt-5 │ │gemini  │ │claude   │
          │       │ Cloud Fns +  │  │gpt-5m│ │2.5-pro │ │sonnet-4 │
          │       │ ruvector-pg  │  │gpt-4o│ │2.5-fl  │ │         │
          │       └──────┬───────┘  │o3    │ │        │ │         │
          │              │          └──────┘ └────────┘ └─────────┘
          │              ▼               Keys from Google Secret Manager
          │  ┌───────────────────────────────────┐
          │  │      Existing Cloud Functions      │
          │  │      (No Changes Required)         │
          │  │                                    │
          │  │  • airtable-agent                  │
          │  │  • db-query-agent                  │
          │  │  • case-manager                    │
          │  │  • simulation-agent                │
          │  │  • workflow-search                 │
          │  └───────────────┬───────────────────┘
          │                  │ VPC (10.128.0.0/20)
          │                  ▼
          │  ┌───────────────────────────────────┐
          │  │     ruvector-postgres VM           │
          └─▶│     10.128.0.2:5432               │
             │     PostgreSQL 17.7 + ruvector    │
             │                                    │
             │  PRIMARY DATA STORE:               │
             │  • workflow_chunks (311 rows)      │
             │  • embeddings (320 vectors, 384d) │
             │  • HNSW index (m=16, ef=64)       │
             │  • Case data, analytics, metrics  │
             └───────────────────────────────────┘

# Bundled MongoDB uses local storage — no connection string needed
# The chat-ui-db image starts MongoDB internally on localhost:27017
MONGODB_URL=mongodb://localhost:27017
MONGODB_DB_NAME=conveyor-chat

# Data persists across container restarts via /data volume
# The chat-ui-db image stores MongoDB data at /data/db

import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { StreamableHTTPServerTransport } from "@modelcontextprotocol/sdk/server/streamableHttp.js";
import express from "express";
import { z } from "zod";

const CLOUD_FUNCTIONS = {
  airtable: "https://airtable-agent-hwqrrwrlna-uc.a.run.app",
  dbQuery:  "https://db-query-agent-hwqrrwrlna-uc.a.run.app",
  caseManager: "https://case-manager-hwqrrwrlna-uc.a.run.app",
  simulation: "https://simulation-agent-hwqrrwrlna-uc.a.run.app",
  workflowSearch: "https://us-central1-new-project-473022.cloudfunctions.net/workflow-search",
};

const server = new McpServer({
  name: "conveyor-tools",
  version: "1.0.0",
});

// Tool: Search workflow documents (vector search via ruvector-postgres)
server.tool(
  "search_workflows",
  "Search CLG workflow procedures, FAQs, and case management steps using semantic search. Returns relevant workflow steps for a given query.",
  {
    query: z.string().describe("Natural language query about workflow procedures"),
    limit: z.number().optional().default(5).describe("Max results to return"),
  },
  async ({ query, limit }) => {
    const resp = await fetch(CLOUD_FUNCTIONS.workflowSearch, {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ action: "search", query, limit }),
    });
    const data = await resp.json();
    return { content: [{ type: "text", text: JSON.stringify(data, null, 2) }] };
  }
);

// Tool: Query database analytics
server.tool(
  "query_database",
  "Run analytics queries against the PostgreSQL database. Supports case metrics, revenue forecasts, and trend analysis.",
  {
    query: z.string().describe("Natural language analytics query"),
    type: z.enum(["metrics", "forecast", "trend", "custom"]).optional().default("metrics"),
  },
  async ({ query, type }) => {
    const resp = await fetch(CLOUD_FUNCTIONS.dbQuery, {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ query, type }),
    });
    const data = await resp.json();
    return { content: [{ type: "text", text: JSON.stringify(data, null, 2) }] };
  }
);

// Tool: Case management operations
server.tool(
  "manage_case",
  "Look up case status, get next steps, list cases, or perform case management operations via Airtable.",
  {
    action: z.enum(["status", "list", "next_steps", "update"]).describe("Case action"),
    caseId: z.string().optional().describe("Case ID (e.g., C-02420)"),
    filters: z.record(z.string()).optional().describe("Filter criteria for list action"),
  },
  async ({ action, caseId, filters }) => {
    const resp = await fetch(CLOUD_FUNCTIONS.caseManager, {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ action, caseId, filters }),
    });
    const data = await resp.json();
    return { content: [{ type: "text", text: JSON.stringify(data, null, 2) }] };
  }
);

// Tool: Run RL simulations
server.tool(
  "run_simulation",
  "Run reinforcement learning strategy simulations for case settlement optimization. Uses Q-learning and Monte Carlo methods.",
  {
    scenario: z.string().describe("Simulation scenario description"),
    episodes: z.number().optional().default(1000).describe("Number of simulation episodes"),
    strategy: z.enum(["q_learning", "monte_carlo", "policy_gradient"]).optional().default("q_learning"),
  },
  async ({ scenario, episodes, strategy }) => {
    const resp = await fetch(CLOUD_FUNCTIONS.simulation, {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ scenario, episodes, strategy }),
    });
    const data = await resp.json();
    return { content: [{ type: "text", text: JSON.stringify(data, null, 2) }] };
  }
);

// Tool: Airtable CRUD
server.tool(
  "airtable_query",
  "Query or update Airtable records. Supports listing cases, clients, carriers, and performing CRUD operations.",
  {
    action: z.enum(["list", "get", "create", "update"]).describe("CRUD action"),
    table: z.string().describe("Airtable table name (e.g., Cases, Clients, Carriers)"),
    recordId: z.string().optional().describe("Record ID for get/update"),
    filters: z.record(z.string()).optional().describe("Filter criteria"),
    fields: z.record(z.unknown()).optional().describe("Fields for create/update"),
  },
  async ({ action, table, recordId, filters, fields }) => {
    const resp = await fetch(CLOUD_FUNCTIONS.airtable, {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ action, table, recordId, filters, fields }),
    });
    const data = await resp.json();
    return { content: [{ type: "text", text: JSON.stringify(data, null, 2) }] };
  }
);

// Express HTTP transport
const app = express();

app.post("/mcp", async (req, res) => {
  const transport = new StreamableHTTPServerTransport("/mcp");
  await server.connect(transport);
  await transport.handleRequest(req, res);
});

app.get("/health", (_, res) => res.json({ status: "ok" }));

app.listen(3001, () => console.log("MCP Bridge running on :3001"));

gcloud run deploy mcp-bridge \
  --source=infrastructure/gcp/mcp-bridge \
  --platform=managed \
  --region=us-central1 \
  --port=3001 \
  --memory=512Mi \
  --cpu=1 \
  --min-instances=0 \
  --max-instances=5 \
  --vpc-connector=conveyor-connector \
  --allow-unauthenticated

URL: https://mcp.airtable.com/v0/mcp
Auth: Bearer ${AIRTABLE_API_KEY}

URL: https://mcp.googleapis.com/v1/drive
Auth: OAuth2 service account or user token

┌─────────────────────────────────────────────────────────────────┐
│                    HF Chat UI — MCP Clients                      │
│                                                                   │
│  ┌─────────────────┐  ┌─────────────────┐  ┌─────────────────┐  │
│  │ Conveyor Bridge  │  │ Airtable MCP   │  │ Google Drive MCP│  │
│  │ (Custom)         │  │ (Official)      │  │ (Google)        │  │
│  │                  │  │                 │  │                 │  │
│  │ • search_wf      │  │ • list_bases   │  │ • search_files  │  │
│  │ • query_db       │  │ • list_tables  │  │ • read_doc      │  │
│  │ • manage_case    │  │ • read_records │  │ • list_folder   │  │
│  │ • run_sim        │  │ • create_record│  │ • read_sheets   │  │
│  │ • airtable_query │  │ • update_record│  │ • get_metadata  │  │
│  │                  │  │ • search       │  │                 │  │
│  └────────┬─────────┘  └───────┬────────┘  └───────┬─────────┘  │
│           │                    │                    │             │
└───────────┼────────────────────┼────────────────────┼─────────────┘
            │                    │                    │
            ▼                    ▼                    ▼
   Cloud Functions +      Airtable API        Google Drive API
   ruvector-postgres      (airtable.com)      (googleapis.com)

MODELS=`[
  {
    "name": "gpt-5.2",
    "id": "gpt-5.2",
    "displayName": "GPT-5.2 (Latest)",
    "description": "OpenAI's latest flagship model. Best for complex reasoning and analysis.",
    "supportsTools": true,
    "parameters": {
      "temperature": 0.7,
      "max_new_tokens": 4096
    },
    "endpoints": [{
      "type": "openai",
      "baseURL": "https://api.openai.com/v1"
    }]
  },
  {
    "name": "gpt-5.2-pro",
    "id": "gpt-5.2-pro",
    "displayName": "GPT-5.2 Pro",
    "description": "Pro tier with extended reasoning. Best for complex case analysis.",
    "supportsTools": true,
    "parameters": {
      "temperature": 0.5,
      "max_new_tokens": 8192
    },
    "endpoints": [{
      "type": "openai",
      "baseURL": "https://api.openai.com/v1"
    }]
  },
  {
    "name": "gpt-5",
    "id": "gpt-5",
    "displayName": "GPT-5",
    "description": "Strong general-purpose reasoning. Good balance of speed and quality.",
    "supportsTools": true,
    "parameters": {
      "temperature": 0.7,
      "max_new_tokens": 4096
    },
    "endpoints": [{
      "type": "openai",
      "baseURL": "https://api.openai.com/v1"
    }]
  },
  {
    "name": "gpt-5-mini",
    "id": "gpt-5-mini",
    "displayName": "GPT-5 Mini",
    "description": "Fast and cost-effective. Great for FAQ lookups and simple workflow queries.",
    "supportsTools": true,
    "parameters": {
      "temperature": 0.7,
      "max_new_tokens": 4096
    },
    "endpoints": [{
      "type": "openai",
      "baseURL": "https://api.openai.com/v1"
    }]
  },
  {
    "name": "gpt-5-nano",
    "id": "gpt-5-nano",
    "displayName": "GPT-5 Nano",
    "description": "Ultra-fast for simple queries. Lowest cost per token.",
    "supportsTools": true,
    "parameters": {
      "temperature": 0.7,
      "max_new_tokens": 2048
    },
    "endpoints": [{
      "type": "openai",
      "baseURL": "https://api.openai.com/v1"
    }]
  },
  {
    "name": "gpt-4o",
    "id": "gpt-4o",
    "displayName": "GPT-4o (Multimodal)",
    "description": "Multimodal model. Upload images of documents, forms, or damage photos.",
    "multimodal": true,
    "supportsTools": true,
    "parameters": {
      "temperature": 0.5,
      "max_new_tokens": 4096
    },
    "endpoints": [{
      "type": "openai",
      "baseURL": "https://api.openai.com/v1"
    }]
  },
  {
    "name": "o3",
    "id": "o3",
    "displayName": "o3 (Reasoning)",
    "description": "Advanced reasoning model. Best for complex legal/financial analysis.",
    "supportsTools": false,
    "parameters": {
      "max_new_tokens": 4096
    },
    "endpoints": [{
      "type": "openai",
      "baseURL": "https://api.openai.com/v1"
    }]
  },
  {
    "name": "gemini-2.5-pro",
    "id": "gemini-2.5-pro",
    "displayName": "Gemini 2.5 Pro (Google)",
    "description": "Google's most capable model. Already used in the existing chat system.",
    "supportsTools": true,
    "parameters": {
      "temperature": 0.7,
      "max_new_tokens": 4096
    },
    "endpoints": [{
      "type": "openai",
      "baseURL": "https://generativelanguage.googleapis.com/v1beta/openai",
      "apiKey": "${GOOGLE_API_KEY}"
    }]
  },
  {
    "name": "gemini-2.5-flash",
    "id": "gemini-2.5-flash",
    "displayName": "Gemini 2.5 Flash (Google)",
    "description": "Google's fast model. Good for quick workflow lookups.",
    "supportsTools": true,
    "parameters": {
      "temperature": 0.7,
      "max_new_tokens": 4096
    },
    "endpoints": [{
      "type": "openai",
      "baseURL": "https://generativelanguage.googleapis.com/v1beta/openai",
      "apiKey": "${GOOGLE_API_KEY}"
    }]
  },
  {
    "name": "claude-sonnet-4",
    "id": "claude-sonnet-4",
    "displayName": "Claude Sonnet 4 (Anthropic)",
    "description": "Anthropic's balanced model. Strong instruction following and coding.",
    "supportsTools": true,
    "parameters": {
      "temperature": 0.7,
      "max_new_tokens": 4096
    },
    "endpoints": [{
      "type": "openai",
      "baseURL": "https://api.anthropic.com/v1",
      "apiKey": "${ANTHROPIC_API_KEY}",
      "defaultHeaders": {
        "anthropic-version": "2023-06-01"
      }
    }]
  }
]`

# All 8 secrets needed for hf-chat-ui
SECRETS=(
  openai-api-key        # GPT-5 models
  anthropic-api-key     # Claude models
  google-api-key        # Gemini models
  airtable-api-key      # Airtable MCP
  airtable-base-id      # Airtable base reference
  google-client-id      # Google OAuth + Drive MCP
  google-client-secret   # Google OAuth + Drive MCP
  gemini-api-key        # Backup Gemini key
)

# Verify all secrets exist
for secret in "${SECRETS[@]}"; do
  echo -n "$secret: "
  gcloud secrets versions access latest --secret="$secret" \
    --project=new-project-473022 2>/dev/null | head -c 12 && echo "... ✓" || echo "MISSING"
done

# Grant access to compute service account
for secret in "${SECRETS[@]}"; do
  gcloud secrets add-iam-policy-binding "$secret" \
    --project=new-project-473022 \
    --member="serviceAccount:[email protected]" \
    --role="roles/secretmanager.secretAccessor" \
    --quiet 2>/dev/null || true
done

# ── Model Provider ──────────────────────────────────────
OPENAI_BASE_URL=https://api.openai.com/v1
# OPENAI_API_KEY injected from Secret Manager

# ── Database ────────────────────────────────────────────
# MONGODB_URL injected from Secret Manager
MONGODB_DB_NAME=conveyor-chat

# ── Branding ────────────────────────────────────────────
PUBLIC_APP_NAME=Conveyor AI
PUBLIC_APP_DESCRIPTION=Insurance Case Management & Revenue Operations Assistant powered by GPT-5
PUBLIC_ORIGIN=https://chat.conveyorclaims.ai

# ── Authentication (Google OAuth) ───────────────────────
OPENID_PROVIDER_URL=https://accounts.google.com
OPENID_CLIENT_ID=245235083640-gkbo4otq57lqeisuigcat0bg037f49oc.apps.googleusercontent.com
# OPENID_CLIENT_SECRET injected from Secret Manager
OPENID_SCOPES=openid profile email
OPENID_NAME_CLAIM=name
COOKIE_SECURE=true
COOKIE_SAMESITE=lax

# ── MCP Tools (3 servers: Custom Bridge + Airtable + Google Drive) ──
MCP_SERVERS=`[
  {
    "name": "Conveyor Tools",
    "description": "Workflow search, DB analytics, case management, simulations via ruvector-postgres and Cloud Functions",
    "url": "https://mcp-bridge-hwqrrwrlna-uc.a.run.app/mcp"
  },
  {
    "name": "Airtable",
    "description": "Direct Airtable base access — browse tables, search records, create/update cases, view schemas",
    "url": "https://mcp.airtable.com/v0/mcp",
    "headers": {
      "Authorization": "Bearer ${AIRTABLE_API_KEY}"
    }
  },
  {
    "name": "Google Drive",
    "description": "Search and read CLG Workflow documents, forms, and templates from Google Drive shared folders",
    "url": "https://mcp.googleapis.com/v1/drive",
    "headers": {
      "Authorization": "Bearer ${GOOGLE_DRIVE_TOKEN}"
    }
  }
]`
MCP_TOOL_TIMEOUT_MS=30000

# ── Smart Router ────────────────────────────────────────
LLM_ROUTER_FALLBACK_MODEL=gpt-5
LLM_ROUTER_ENABLE_TOOLS=true
LLM_ROUTER_TOOLS_MODEL=gpt-5.2
PUBLIC_LLM_ROUTER_DISPLAY_NAME=Auto (Omni)
PUBLIC_LLM_ROUTER_ALIAS_ID=omni

# ── Voice ───────────────────────────────────────────────
TRANSCRIPTION_MODEL=openai/whisper-large-v3-turbo

# ── Web Search ──────────────────────────────────────────
USE_LOCAL_WEBSEARCH=true

# ── Features ────────────────────────────────────────────
LLM_SUMMARIZATION=true
ENABLE_DATA_EXPORT=true
ALLOW_IFRAME=false

# ── Rate Limits ─────────────────────────────────────────
USAGE_LIMITS={"messagesPerMinute": 20, "conversations": 100, "tools": 50}

# ── System Prompt (Conveyor Identity) ───────────────────
TASK_MODEL=gpt-5-mini

steps:
  # Step 1: Pull the pre-built HuggingFace Chat UI image
  - name: 'gcr.io/cloud-builders/docker'
    args: ['pull', 'ghcr.io/huggingface/chat-ui:latest']

  # Step 2: Tag for GCR
  - name: 'gcr.io/cloud-builders/docker'
    args: [
      'tag',
      'ghcr.io/huggingface/chat-ui:latest',
      'gcr.io/${PROJECT_ID}/hf-chat-ui:${_VERSION}'
    ]

  # Step 3: Push versioned tag
  - name: 'gcr.io/cloud-builders/docker'
    args: ['push', 'gcr.io/${PROJECT_ID}/hf-chat-ui:${_VERSION}']

  # Step 4: Push latest tag
  - name: 'gcr.io/cloud-builders/docker'
    args: [
      'tag',
      'gcr.io/${PROJECT_ID}/hf-chat-ui:${_VERSION}',
      'gcr.io/${PROJECT_ID}/hf-chat-ui:latest'
    ]
  - name: 'gcr.io/cloud-builders/docker'
    args: ['push', 'gcr.io/${PROJECT_ID}/hf-chat-ui:latest']

  # Step 5: Deploy to Cloud Run
  - name: 'gcr.io/google.com/cloudsdktool/cloud-sdk'
    entrypoint: gcloud
    args: [
      'run', 'deploy', 'hf-chat-ui',
      '--image', 'gcr.io/${PROJECT_ID}/hf-chat-ui:${_VERSION}',
      '--platform', 'managed',
      '--region', 'us-central1',
      '--port', '3000',
      '--memory', '2Gi',
      '--cpu', '2',
      '--min-instances', '0',
      '--max-instances', '10',
      '--timeout', '300',
      '--vpc-connector', 'conveyor-connector',
      '--allow-unauthenticated',
      '--set-env-vars', 'OPENAI_BASE_URL=https://api.openai.com/v1,MONGODB_DB_NAME=conveyor-chat,PUBLIC_APP_NAME=Conveyor AI,PUBLIC_ORIGIN=https://chat.conveyorclaims.ai,LLM_SUMMARIZATION=true,ENABLE_DATA_EXPORT=true',
      '--set-secrets', 'OPENAI_API_KEY=openai-api-key:latest,ANTHROPIC_API_KEY=anthropic-api-key:latest,GOOGLE_API_KEY=google-api-key:latest,AIRTABLE_API_KEY=airtable-api-key:latest,GOOGLE_CLIENT_ID=google-client-id:latest,GOOGLE_CLIENT_SECRET=google-client-secret:latest',
    ]

substitutions:
  _VERSION: 'v1'

options:
  logging: CLOUD_LOGGING_ONLY
timeout: 600s

# Verify domain ownership (one-time)
gcloud domains verify conveyorclaims.ai --project=new-project-473022

# Map custom domain to the Cloud Run service
gcloud run domain-mappings create \
  --service=hf-chat-ui \
  --domain=chat.conveyorclaims.ai \
  --region=us-central1 \
  --project=new-project-473022

Console → APIs & Services → Credentials → OAuth 2.0 Client ID
→ Authorized redirect URIs → Add:
   https://chat.conveyorclaims.ai/login/callback

{
  "name": "Conveyor AI",
  "preprompt": "You are Conveyor AI, an Insurance Case Management & Revenue Operations Assistant for CLG (Claims Litigation Group).\n\n## Your Capabilities\n- Case management: Look up case status, next steps, due dates, assigned roles\n- Workflow guidance: Step-by-step procedures from CLG workflow documents\n- Revenue forecasting: Analytics and trend analysis\n- Strategy optimization: RL-based settlement strategy simulations\n- Airtable operations: Query and update case records\n\n## Response Style\n- Start conversationally: 'Great question —', 'Yes —', 'Got it —'\n- Use emoji markers: ✅ ❌ ⚠️ 🔑 💰 📌 for scannability\n- Bold field names: **Next Steps**, **Case Status**, **RS Due Date**\n- End with a key takeaway: 🔑 or 🧠 summary\n- Offer proactive follow-up: 'If you want, I can also...'\n- NEVER expose: similarity scores, chunk IDs, function names, JSON, silo numbers\n- ALWAYS attribute sources by document name: 'Referrals Workflow', 'FAQ's'\n\n## Available Tools\nYou have access to Conveyor Tools via MCP. Use them to:\n- search_workflows: Search CLG workflow procedures and FAQs\n- query_database: Run analytics against PostgreSQL\n- manage_case: Look up or update case status via Airtable\n- run_simulation: Run RL strategy simulations\n- airtable_query: Direct Airtable CRUD operations",
  "model": "gpt-5.2"
}

# 1. Deploy Chat UI to Cloud Run (bundled MongoDB sidecar via chat-ui-db image)
gcloud run deploy hf-chat-ui \
  --image=ghcr.io/huggingface/chat-ui-db:latest \
  --platform=managed \
  --region=us-central1 \
  --port=3000 \
  --memory=2Gi \
  --cpu=2 \
  --min-instances=1 \
  --max-instances=10 \
  --timeout=300 \
  --vpc-connector=conveyor-connector \
  --allow-unauthenticated \
  --set-env-vars="OPENAI_BASE_URL=https://api.openai.com/v1,MONGODB_URL=mongodb://localhost:27017,MONGODB_DB_NAME=conveyor-chat,PUBLIC_APP_NAME=Conveyor AI,PUBLIC_ORIGIN=https://chat.conveyorclaims.ai,LLM_SUMMARIZATION=true,ENABLE_DATA_EXPORT=true,ALLOW_IFRAME=false,USE_LOCAL_WEBSEARCH=true" \
  --set-secrets="OPENAI_API_KEY=openai-api-key:latest,ANTHROPIC_API_KEY=anthropic-api-key:latest,GOOGLE_API_KEY=google-api-key:latest,AIRTABLE_API_KEY=airtable-api-key:latest,GOOGLE_CLIENT_ID=google-client-id:latest,GOOGLE_CLIENT_SECRET=google-client-secret:latest" \
  --project=new-project-473022

# 2. Deploy MCP Bridge (connects Chat UI tools to existing Cloud Functions + ruvector-postgres)
gcloud run deploy mcp-bridge \
  --source=infrastructure/gcp/mcp-bridge \
  --platform=managed \
  --region=us-central1 \
  --port=3001 \
  --memory=512Mi \
  --cpu=1 \
  --vpc-connector=conveyor-connector \
  --allow-unauthenticated \
  --project=new-project-473022

# 3. Map custom domain
gcloud run domain-mappings create \
  --service=hf-chat-ui \
  --domain=chat.conveyorclaims.ai \
  --region=us-central1 \
  --project=new-project-473022

# 4. Add DNS CNAME record at registrar
# chat.conveyorclaims.ai → ghs.googlehosted.com.

┌──────────────────────────────────────────────────────────────────────────────────┐
│                          GOOGLE CLOUD PLATFORM                                    │
│                          Project: new-project-473022                              │
├──────────────────────────────────────────────────────────────────────────────────┤
│                                                                                   │
│  ┌─────────────────────────────────────────────────────────────────────────────┐  │
│  │                       VPC Network (conveyor-vpc)                             │  │
│  │                                                                              │  │
│  │  ┌─────────────────────────────────────────────────────────────┐             │  │
│  │  │                    Cloud Run Services                        │             │  │
│  │  │                                                              │             │  │
│  │  │  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐      │             │  │
│  │  │  │  hf-chat-ui  │  │ chat-system  │  │  mcp-bridge  │      │             │  │
│  │  │  │  (NEW)       │  │ (existing)   │  │  (NEW)       │      │             │  │
│  │  │  │              │  │              │  │              │      │             │  │
│  │  │  │ SvelteKit    │  │ React+Vite   │  │ MCP Server   │      │             │  │
│  │  │  │ GPT-5 models │  │ Gemini       │  │ Tool bridge  │      │             │  │
│  │  │  │ Port 3000    │  │ Port 8080    │  │ Port 3001    │      │             │  │
│  │  │  └──────┬───────┘  └──────────────┘  └──────┬───────┘      │             │  │
│  │  │         │                                     │              │             │  │
│  │  │         │chat.conveyorclaims.ai               │              │             │  │
│  │  └─────────┼─────────────────────────────────────┼──────────────┘             │  │
│  │            │                                     │                            │  │
│  │  ┌────────┼─────────────────────────────────────┼───────────────────┐        │  │
│  │  │        │         Cloud Functions              │                   │        │  │
│  │  │        │                                      │                   │        │  │
│  │  │        │  • airtable-agent  ◄─────────────────┤                   │        │  │
│  │  │        │  • db-query-agent  ◄─────────────────┤                   │        │  │
│  │  │        │  • case-manager    ◄─────────────────┤                   │        │  │
│  │  │        │  • simulation-agent◄─────────────────┤                   │        │  │
│  │  │        │  • workflow-search ◄─────────────────┘                   │        │  │
│  │  │        │                                                          │        │  │
│  │  └────────┼──────────────────────────────────────────────────────────┘        │  │
│  │           │                                                                   │  │
│  │  ┌────────▼─────────┐                                                         │  │
│  │  │  ruvector-postgres│                                                        │  │
│  │  │  10.128.0.2:5432 │                                                        │  │
│  │  │  PostgreSQL 17.7  │                                                        │  │
│  │  │  ruvector 2.0.1   │                                                        │  │
│  │  └──────────────────┘                                                         │  │
│  └───────────────────────────────────────────────────────────────────────────────┘  │
│                                                                                     │
│  ┌───────────────────────────┐    ┌───────────────────────────────────┐              │
│  │  Secret Manager           │    │  AI Providers (Multi-Provider)    │              │
│  │  • openai-api-key         │    │  • OpenAI    → GPT-5 family      │              │
│  │  • anthropic-api-key      │    │  • Google    → Gemini 2.5        │              │
│  │  • google-api-key         │    │  • Anthropic → Claude Sonnet 4   │              │
│  │  • airtable-api-key       │    └───────────────────────────────────┘              │
│  │  • ruvector-db-password   │                                                       │
│  └───────────────────────────┘                                                       │
└─────────────────────────────────────────────────────────────────────────────────────┘

OPENID_PROVIDER_URL=https://accounts.google.com
OPENID_CLIENT_ID=245235083640-gkbo4otq57lqeisuigcat0bg037f49oc.apps.googleusercontent.com
OPENID_SCOPES=openid profile email
OPENID_NAME_CLAIM=name
COOKIE_SECURE=true
COOKIE_SAMESITE=lax