Iris Research (MCP on Cloud Run)

Iris Research is a stateless AI assistant hosted on Google Cloud Run. It accepts a phone number, scrapes real-time news using Apify, generates a voice-ready prompt using a custom LLM flow, and places an outbound call using Vapi — all through a single MCP-compatible HTTP endpoint.

---

🧠 Overview

This is a single MCP tool named iris-research that:

1. Accepts a phone number and research query.
2. Scrapes relevant news using Apify.
3. Generates a summarized spoken response using a language model.
4. Places a call to the number using Vapi and reads the result aloud.

---

🌐 MCP Tool

POST /mcp

Request:

{
  "jsonrpc": "2.0",
  "method": "iris-research",
  "params": {
    "phoneNumber": "+14155550123",
    "query": "latest cybersecurity news"
  },
  "id": 1
}

Response:

{
  "jsonrpc": "2.0",
  "result": {
    "content": [
      {
        "type": "text",
        "text": "Call placed to +14155550123 with a summary of the latest cybersecurity news."
      }
    ]
  },
  "id": 1
}

This endpoint is fully MCP-compatible and can be called by another MCP agent, Vapi, or any HTTP/JSON system.

---

☁️ Deploying to Google Cloud Run

1. Prerequisites

• Node.js 18+ (locally)
• GCP project with Cloud Run enabled
• Docker & gcloud CLI installed and authenticated

2. Build & Deploy

gcloud builds submit --tag gcr.io/YOUR_PROJECT_ID/iris-research

gcloud run deploy iris-research \
  --image gcr.io/YOUR_PROJECT_ID/iris-research \
  --platform managed \
  --region us-central1 \
  --allow-unauthenticated

Replace YOUR_PROJECT_ID with your actual GCP project ID.

3. Test

curl -X POST https://YOUR_RUN_URL/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "method": "iris-research",
    "params": {
      "phoneNumber": "+14155550123",
      "query": "latest AI news"
    },
    "id": 1
  }'

---

📁 Project Structure

iris-research/
├── src/
│   ├── index.ts             # Express server with stateless MCP endpoint
│   ├── research.ts          # Handles orchestration of scraping, prompting, and calling
│   ├── getNewsData.ts       # Apify integration
│   ├── generatePrompt.ts    # LLM prompt creation
│   └── makeOutboundCall.ts  # Vapi outbound call integration
├── package.json
├── tsconfig.json
├── Dockerfile
└── .dockerignore

---

🔐 Environment Variables

For local testing, create a .env file or set the following:

APIFY_API_TOKEN=...
VAPI_API_KEY=...
GROQ_API_KEY=...
PHONE_NUMBER=...

---

⚙️ Tech Stack

• TypeScript
• Node.js 20+ (ESM + .js imports)
• MCP SDK
• Apify
• Groq or OpenAI
• Vapi
• Google Cloud Run

---

👥 Contributors

• Shapor Naghibzadeh
• Tom Lyttelton

---

📄 License

MIT License