Cost Transparency

Track and control costs for every MCP agent in real-time.

Real-time Cost Tracking

Every tool invocation includes cost information:

{
  "tool": "llm/chat",
  "response": {...},
  "cost": {
    "input_tokens": 150,
    "output_tokens": 200,
    "total": 0.0035
  }
}

Session Costs

Track costs per agent session:

curl https://api.gateflow.ai/v1/mcp/sessions/sess_abc123/cost \
  -H "Authorization: Bearer gw_prod_..."

Response:

{
  "session_id": "sess_abc123",
  "agent_id": "agent_support_bot",
  "started_at": "2024-01-15T10:00:00Z",
  "cost": {
    "total": 0.45,
    "by_tool": {
      "llm/chat": 0.40,
      "retrieval/search": 0.05
    }
  },
  "limits": {
    "session_limit": 5.00,
    "remaining": 4.55
  }
}

Cost Limits

Per-Session Limits

permissions:
  rate_limits:
    cost_per_session: 5.00  # USD

When reached:

{
  "error": "session_cost_limit_exceeded",
  "message": "Session cost limit of $5.00 reached",
  "spent": 5.02
}

Daily Limits

permissions:
  rate_limits:
    cost_per_day: 100.00  # USD

Monthly Limits

permissions:
  rate_limits:
    cost_per_month: 2000.00  # USD

Cost Alerts

Configure alerts before limits are hit:

curl -X POST https://api.gateflow.ai/v1/mcp/alerts \
  -H "Authorization: Bearer gw_prod_..." \
  -H "Content-Type: application/json" \
  -d '{
    "agent_id": "agent_123",
    "condition": {
      "metric": "session_cost",
      "threshold": 4.00
    },
    "notify": ["slack"]
  }'

Cost Analytics

By Agent

curl https://api.gateflow.ai/v1/mcp/analytics/cost \
  -H "Authorization: Bearer gw_prod_..." \
  -G -d "group_by=agent" -d "period=30d"

By Tool

curl https://api.gateflow.ai/v1/mcp/analytics/cost \
  -H "Authorization: Bearer gw_prod_..." \
  -G -d "group_by=tool" -d "period=30d"

Agent Self-Inspection

Agents can check their own costs:

# Using the whoami tool
my_usage = await agent.tools.whoami()
print(f"Session cost: ${my_usage.session_cost}")
print(f"Remaining budget: ${my_usage.cost_remaining}")

Best Practices

1. Set conservative limits initially, increase as needed
2. Use alerts at 80% of limits
3. Review costs weekly to identify optimization opportunities
4. Use cheaper models for simple tasks

Next Steps

• Rate Limits - Request rate limiting
• Audit Logging - Track all agent activity