Cloudflare Cost Report

Generate a comprehensive cost analysis for your Cloudflare architecture with optimization recommendations.

Arguments: "$ARGUMENTS"

Modes

Mode	Description	MCP Tools
Static (default)	Analyze config and code only	None
Validate (--validate)	Compare static findings against live data	observability, ai-gateway, bindings

Workflow

Step 0: Parse Arguments and Detect Mode

Parse $ARGUMENTS for:

• --validate: Enable live data validation
• --detailed: Per-binding cost breakdown
• [path]: Explicit wrangler config path

Mode Selection:

• If --validate present: Live Validation Mode
• Otherwise: Static Analysis Mode

Step 1: MCP Availability Check (if --validate)

Before using MCP tools, verify availability:

// Lightweight probe to test MCP connectivity
mcp__cloudflare-bindings__workers_list()

Expected outcomes:

• Success: MCP tools available, proceed with live validation
• Failure/Timeout: Note "MCP tools unavailable" in output
  • Continue with static analysis
  • Tag all findings as [STATIC]
  • Add footer: "Note: Some findings may be incomplete without live observability data"

Step 2: Locate Wrangler Config

If path provided, use it. Otherwise search:

# Find wrangler.toml or wrangler.jsonc
find . -name "wrangler.toml" -o -name "wrangler.jsonc" | head -5

Step 3: Parse Configuration

Read the wrangler config and identify all bindings:

• D1 databases
• R2 buckets
• KV namespaces
• Queues (check max_retries)
• Vectorize indexes
• AI bindings
• Durable Objects
• Analytics Engine datasets

Step 4: Gather Usage Data

Static Mode

Analyze code patterns:

• Count D1 queries in source files
• Identify batch vs per-row inserts
• Check queue publish frequency
• Scan for AI model usage

Validate Mode (MCP Available)

Use MCP tools to collect real usage metrics. Reference @skills/probes/SKILL.md for query patterns.

1. Worker Metrics (cloudflare-observability):

   mcp__cloudflare-observability__query_worker_observability({
     view: "calculations",
     parameters: {
       calculations: [
         { operator: "count", as: "total_requests" },
         { operator: "sum", field: "$metadata.cpuTime", as: "total_cpu_ms" }
       ],
       groupBys: [{ type: "string", value: "$metadata.service" }]
     },
     timeframe: { reference: "now", offset: "-30d" }
   })
   

2. AI Gateway Logs (cloudflare-ai-gateway):

   mcp__cloudflare-ai-gateway__list_logs({
     gateway_id: "...",
     per_page: 1000
   })
   // Aggregate by model, calculate actual costs
   

3. Resource Lists (cloudflare-bindings):

   mcp__cloudflare-bindings__d1_databases_list()
   mcp__cloudflare-bindings__r2_buckets_list()
   mcp__cloudflare-bindings__kv_namespaces_list()
   

Step 5: Calculate Costs

Apply 2026 Cloudflare pricing:

Service	Read	Write	Storage
D1	$0.25/B rows	$1.00/M rows	$0.75/GB
R2	$0.36/M (Class B)	$4.50/M (Class A)	$0.015/GB
KV	$0.50/M	$5.00/M	$0.50/GB
Queues	$0.40/M messages	-	-
Workers	$0.30/M requests	-	-
Workers AI	Varies by model	-	-

Step 6: Identify Optimizations

Check for common cost traps:

1. D1 Anti-patterns:

   • Per-row inserts (should batch)
   • Missing indexes (SCAN TABLE)
   • SELECT * usage

2. Queue Anti-patterns:

   • max_retries > 1 for idempotent consumers
   • Missing dead-letter queue

3. AI Anti-patterns:

   • No caching for repeated prompts
   • Large models for simple tasks
   • Missing rate limiting

4. General Anti-patterns:

   • Frequent cron jobs (batch work)
   • Missing Smart Placement
   • Subrequest exhaustion

Step 7: Live Validation (if --validate and MCP available)

Compare static estimates against live data:

1. D1 Write Volume

   • Static estimate: Count INSERT/UPDATE in code × estimated frequency
   • Live data: Query observability for actual write counts
   • Compare and tag: [LIVE-VALIDATED] or [LIVE-REFUTED]

2. AI Gateway Costs

   • Static estimate: Model pricing × estimated requests
   • Live data: Aggregate from AI Gateway logs
   • Identify cache hit opportunities

3. Queue Retry Rates

   • Static estimate: Based on max_retries config
   • Live data: Query actual retry patterns
   • Calculate real retry cost multiplier

4. EXPLAIN QUERY PLAN (D1)

   mcp__cloudflare-bindings__d1_database_query({
     database_id: "...",
     sql: "EXPLAIN QUERY PLAN [detected query]"
   })
   

   • Validate index usage claims
   • Identify SCAN TABLE issues not caught statically

Step 8: Generate Report with Provenance Tags

Output format:

# Cloudflare Cost Report
Generated: [date]
Project: [wrangler name]
Mode: [Static | Live Validated]

## Summary
- **Estimated Monthly Cost**: $XX.XX
- **Cost Trend**: [Stable|Rising|Declining]
- **Optimization Potential**: $XX.XX/month
- **Validation Status**: [Full | Partial | Static Only]

## Service Breakdown

| Service | Monthly Cost | % Total | Status | Source |
|---------|-------------|---------|--------|--------|
| D1 | $XX.XX | XX% | [OK|Warning|Critical] | [STATIC|LIVE] |
| Workers | $XX.XX | XX% | [OK|Warning|Critical] | [STATIC|LIVE] |
| ... | ... | ... | ... | ... |

## Cost Drivers

### [LIVE-VALIDATED] D1 Writes - XX% of total
- **Static Estimate**: $XX.XX/month
- **Live Actual**: $XX.XX/month
- **Evidence**: Observability data, 30-day window
- Root cause: [explanation]

### [STATIC] Queue Retries - XX% of total
- **Estimate**: $XX.XX/month
- Root cause: [explanation]
- Note: Live validation unavailable

### [LIVE-REFUTED] AI Gateway - XX% of total
- **Static Estimate**: $XX.XX/month (based on code patterns)
- **Live Actual**: $XX.XX/month (cache hits reducing cost)
- **Evidence**: AI Gateway logs show 40% cache hit rate

## Optimization Opportunities

| Opportunity | Savings | Effort | Priority | Source |
|-------------|---------|--------|----------|--------|
| [Description] | $XX/mo | [Low|Med|High] | [P1|P2|P3] | [STATIC|LIVE-VALIDATED] |

## Warnings

- [List any concerning patterns with provenance tags]

## Action Items

1. [ ] [LIVE-VALIDATED] [Specific action with file:line reference]
2. [ ] [STATIC] [Specific action with file:line reference]

---
**Finding Tags:**
- `[STATIC]` - Inferred from code/config analysis
- `[LIVE-VALIDATED]` - Confirmed by observability (--validate mode)
- `[LIVE-REFUTED]` - Code smell not observed in production
- `[INCOMPLETE]` - Some MCP tools unavailable

Usage Examples

Basic cost report (static):

/cf-costs

Cost report with live validation:

/cf-costs --validate

Specific wrangler file with validation:

/cf-costs workers/wrangler.jsonc --validate

Detailed report with per-binding breakdown:

/cf-costs --detailed --validate

Tips

• Run weekly to track cost trends
• Use --validate for accurate cost attribution
• Compare before/after deployments
• Use --detailed for capacity planning
• Check D1 writes first (usually the culprit)
• If MCP tools unavailable, findings marked [STATIC] should be verified manually

MCP Tool Requirements

For --validate mode, these MCP servers should be configured:

• cloudflare-observability - Worker metrics, D1 usage
• cloudflare-ai-gateway - AI cost data
• cloudflare-bindings - Resource lists, D1 queries

If any MCP server is unavailable, the command will:

1. Note which tools are missing
2. Continue with available data
3. Tag affected findings as [INCOMPLETE]