# TOON Output Format
## What is TOON?
**Token-Oriented Object Notation** is a compact, LLM-friendly format that minimizes token usage while maintaining readability.
Reference: https://github.com/toon-format/toon
## Why TOON?
- ~40% fewer tokens than JSON
- Combines YAML indentation with CSV-style tabular arrays
- Lossless round-trip conversion
- Optimized for uniform object arrays (common in refactoring output)
## Syntax Examples
### Simple Object
```
preview:
refactoring: extract-method
status: success
files_affected: 1
```
### Tabular Array (uniform objects)
```
refactorings[3]{name,category,params,description}:
extract-method,composing_methods,"target,name",Extract code block into new method
rename-method,simplifying_method_calls,"target,new_name",Rename a method
move-method,moving_features,"target,to_class",Move method to another class
```
### Mixed Output
```
analysis:
path: src/order.py
language: python
backend_supported: true
smells[2]{type,location,severity,suggested_refactoring,details}:
long-method,Order::calculate_total,high,extract-method,"45 lines (threshold: 20)"
feature-envy,Order::format_address,medium,move-method,"Accesses Customer 8 times"
```
## Implementation
The encoder is in `utils/toon.py`:
```python
from mcp_refactoring.utils import encode_toon
result = encode_toon({
"status": "success",
"files": [
{"path": "order.py", "action": "modified"},
{"path": "customer.py", "action": "modified"},
]
})
```
## Backend Requirements
Language backends should support `--output=toon` flag to return TOON-formatted output directly, avoiding parsing overhead in the MCP server.
