IntegrityPulse FinOps

Stop your AI agents from hallucinating cloud costs. Get real pricing forecasts inside the conversation.

---

What is this?

IntegrityPulse FinOps is a remote MCP server that gives AI coding agents accurate, real-time cloud deployment cost forecasts. Instead of your agent guessing that "an EC2 instance costs around $50/month," it calls a tool backed by a verified pricing matrix and returns a line-item breakdown.

One tool. Three providers. Zero hallucinations.

User: "What would it cost to run our API on AWS with an m5.large, a managed Postgres, and Redis?"

Agent (via IntegrityPulse FinOps):
  | Service                    | Category | Hours | Est. Cost |
  |----------------------------|----------|-------|-----------|
  | m5.large                   | Compute  | 730   | $70.08    |
  | rds.postgres.db.m5.large   | Database | 730   | $204.40   |
  | elasticache.redis.t3.micro | Cache    | 730   | $11.68    |

  Total Estimated Monthly Cost: $286.16

---

Quick Start

Claude Desktop

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "integritypulse": {
      "type": "streamable-http",
      "url": "https://integritypulse.marywomack.workers.dev/mcp",
      "headers": {
        "x-api-key": "YOUR_API_KEY"
      }
    }
  }
}

Cursor

Add to your .cursor/mcp.json:

{
  "mcpServers": {
    "integritypulse": {
      "type": "streamable-http",
      "url": "https://integritypulse.marywomack.workers.dev/mcp",
      "headers": {
        "x-api-key": "YOUR_API_KEY"
      }
    }
  }
}

cURL

curl -X POST https://integritypulse.marywomack.workers.dev/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "x-api-key: YOUR_API_KEY" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "forecast_deployment_cost",
      "arguments": {
        "provider": "AWS",
        "services_to_add": [
          {"service_name": "m5.large", "estimated_usage_hours": 730}
        ]
      }
    }
  }'

---

The Tool: forecast_deployment_cost

Parameter	Type	Description
provider	"AWS" | "GCP" | "AZURE"	Cloud provider to price against
services_to_add	Array<{ service_name, estimated_usage_hours }>	Services to include in the forecast

Supported services:

AWS	GCP	Azure
t3.micro, t3.medium, m5.large	e2-micro, e2-medium, n2-standard-2	B1s, B2s, D2s_v3
rds.postgres.db.t3.micro, rds.postgres.db.m5.large	cloudsql.postgres.db-custom-1-3840, cloudsql.postgres.db-custom-4-15360	postgresql.flexible.b1ms
elasticache.redis.t3.micro	memorystore.redis.1gb
s3.standard.1tb

---

Why It Exists: FinOps for the Agentic Era

AI agents are moving from "write me code" to "deploy this for me." When an agent provisions infrastructure, cost accuracy isn't a nice-to-have -- it's a financial control.

The problem: LLMs hallucinate pricing. They confidently tell you an RDS instance costs "$15/month" when the real number is $204. When agents start executing deployments autonomously, these hallucinations become real invoices.

The solution: IntegrityPulse FinOps is a grounded pricing oracle that agents call as a tool. The pricing matrix is maintained, versioned, and deterministic. No generation, no guessing.

Agentic Commerce & the Revenue Gate

This server is also a reference implementation of Revenue-Gated MCP -- a pattern for monetizing MCP tools without breaking the agent experience.

Here's how it works:

1. Every request carries an x-api-key header.
2. The server checks the user's tier and monthly_usage_count in Supabase.
3. Free tier users get 25 operations/month.
4. When the limit is hit, the server does NOT return an HTTP error. Instead, it returns a valid MCP tool result with isError: true:

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [{
      "type": "text",
      "text": "IntegrityPulse FinOps Alert: Your free monthly tier (25/25 operations) has been exhausted. To generate this architectural cost forecast, please upgrade to the Pro tier here: https://billing.openclaw.com/upgrade. Once upgraded, ask me to retry."
    }],
    "isError": true
  }
}

Why isError: true instead of HTTP 402/429?

Because the consumer is an LLM, not a browser. An HTTP error gets swallowed by the MCP transport layer and the user never sees it. By returning a tool result with isError: true, the upgrade CTA lands directly in the conversation where the user is already engaged. The agent reads it, surfaces it, and the user can act on it immediately. This is commerce at the point of intent.

Stripe Integration

When a user upgrades via the billing link, Stripe fires a checkout.session.completed webhook to /api/webhook/stripe. The server automatically:

• Updates their tier from FREE to PRO
• Resets their usage counter to 0
• The next agent request goes through seamlessly

---

Architecture

                         +-----------------+
  Claude / Cursor        |  Cloudflare     |       +------------+
  (MCP Client)  -------->|  Worker (Hono)  |------>|  Supabase  |
                POST     |                 |       |  (users)   |
                /mcp     |  Revenue Gate   |       +------------+
                         |  MCP Server     |
                         |  Stripe Webhook |------>  Stripe API
                         +-----------------+

• Runtime: Cloudflare Workers (edge, ~0ms cold start)
• Framework: Hono (lightweight, Workers-native)
• MCP SDK: @modelcontextprotocol/sdk with Streamable HTTP transport
• Auth: API key in x-api-key or Authorization: Bearer header
• Database: Supabase (PostgreSQL)
• Payments: Stripe Checkout + Webhooks

---

Self-Hosting

git clone https://github.com/maryadawson-code/integritypulse.git
cd integritypulse
npm install

# Configure secrets
cp .env.example .env
# Edit .env with your Supabase and Stripe credentials

# Local development
npx wrangler dev

# Deploy to Cloudflare Workers
npx wrangler secret put SUPABASE_URL
npx wrangler secret put SUPABASE_SERVICE_KEY
npx wrangler secret put STRIPE_SECRET_KEY
npx wrangler secret put STRIPE_WEBHOOK_SECRET
npx wrangler deploy

---

License

MIT