# Token Economics
## Traditional vs Meta-MCP
```mermaid
flowchart TB
subgraph Traditional["Traditional"]
T1[Start] --> T2[Load ALL tools]
T2 --> T3["16k-48k+ tokens"]
end
subgraph Meta["Meta-MCP"]
M1[Start] --> M2["1.9k startup"]
M2 --> M3["only load what you use"]
M3 --> M4["~3k typical"]
end
style T3 fill:#ff6b6b
style M4 fill:#51cf66
```
## Savings
Traditional loads ALL available tools at startup. Meta-MCP only loads what you use.
| Servers | Tools Available | Traditional | You Use | Meta-MCP | Savings |
|---------|-----------------|-------------|---------|----------|---------|
| 1 | 25 | 16,000 | 2 | 3,200 | **80%** |
| 3 | 75 | 48,000 | 2 | 3,200 | **93%** |
| 3 | 75 | 48,000 | 5 | 5,100 | **89%** |
Formula: `1,900 + (tools used × 640)`
## Request Flow
```mermaid
sequenceDiagram
participant AI
participant Meta
participant Backend
rect rgb(230,245,255)
Note over AI: Startup: 1.9k tokens
end
AI->>Meta: list_servers()
Meta-->>AI: ["jira", "slack"]
AI->>Meta: get_server_tools(jira, summary_only)
Meta-->>AI: [summaries]
rect rgb(232,245,233)
Note over AI: +640 per tool
end
AI->>Meta: call_tool(jira, search, {})
Meta->>Backend: execute
Backend-->>AI: results
```
