Convert JSON to TOON (Token-Oriented Object Notation) and reduce your LLM token costs by up to 60%. Optimize for ChatGPT, Claude, and all major AI models.
Built for AI developers, prompt engineers, and anyone working with Large Language Models.
TOON (Token-Oriented Object Notation) is a compact, human-readable format designed specifically for Large Language Models (LLMs). It represents the same data as JSON but uses significantly fewer tokens, reducing API costs and improving AI model performance.
Think of TOON as a translation layer: you use JSON programmatically in your applications, then encode it as TOON when sending data to AI models like Claude, ChatGPT, Gemini, or any LLM API.
TOON combines YAML's indentation-based structure for nested objects with a CSV-style tabular layout for uniform arrays. This makes it incredibly efficient for the most common LLM use case: arrays of objects with consistent structure.
{
"users": [
{
"id": 1,
"name": "Alice",
"role": "admin"
},
{
"id": 2,
"name": "Bob",
"role": "user"
}
]
}users[2]{id,name,role}:
1,Alice,admin
2,Bob,userTOON is production-ready and used by developers worldwide to optimize AI applications
TOON uses significantly fewer tokens than JSON, directly reducing your LLM API costs. For high-volume applications, this translates to thousands of dollars in monthly savings.
Fewer tokens mean faster processing times. Your AI applications respond quicker, improving user experience and enabling real-time interactions.
TOON's explicit structure (array lengths, field headers) helps LLMs parse data more reliably, reducing hallucinations and improving response quality.
With 40-60% token savings, you can include more records in the same context window, enabling richer AI responses and better decision-making.
TOON is a lossless representation of JSON. Convert back and forth without data loss. Use JSON in your code, TOON for AI prompts.
Works with all major LLMs. TOON is model-agnostic and future-proof - an efficient way to incorporate structured data into AI prompts.
Based on comprehensive benchmarks across Claude Haiku, Gemini Flash, GPT-5 Nano, and Grok models.View full benchmarks →
Let's discuss how TOON and advanced AI optimization can reduce your costs and improve performance.
Maximize efficiency and accuracy when using TOON with AI agents and the Model Context Protocol
Model Context Protocol (MCP) is an open standard that enables AI applications to securely connect to data sources and tools. It's the backbone of modern AI agent systems, allowing LLMs to access external data, execute functions, and maintain context across interactions.
MCP is used by Claude Desktop, ChatGPT plugins, custom AI agents, and enterprise AI systems to provide structured, reliable data exchange between AI models and external systems.
The key insight: MCP tool inputs stay as JSON (required by the MCP protocol and APIs), but tool outputs can be TOON through server-side conversion.
Why this works: MCP protocol allows text responses, and TOON is just text. LLMs parse TOON naturally (often better than JSON), so you get massive token savings without breaking compatibility.
Real-world token savings from production implementation
Based on TOON benchmarks for Instantly.ai use cases:
| Tool | Token Reduction |
|---|---|
list_accounts | 60.7% 🔥 |
list_campaigns | 50-60% |
list_leads | 50-60% |
get_campaign_analytics | 74.6% 🔥 |
list_emails | 15-30% |
Based on this test, TOON will provide massive savings for:
get_campaign_analytics - 74% savings potentialget_daily_campaign_analytics - 74% savings potentialget_warmup_analytics - 74% savings potentialSelect only essential fields: {email, status, warmup_status, daily_limit}
This would make data uniform enough for TOON
Example 1: Email Accounts
Example 2: Campaign Analytics (30 days)
MCP tools that return lists of accounts, campaigns, or leads can reduce token costs by 50-60% per response. For high-volume AI agents, this translates to thousands of dollars in monthly savings.
AI agents that call multiple MCP tools (list accounts → get analytics → list emails) compound savings across the entire workflow, reducing costs by thousands of tokens per conversation.
With 60% token savings, you can return 2.5x more records in the same context window. Return 250 accounts instead of 100, enabling richer AI responses and better decision-making.
Based on typical LLM pricing at $0.03/1K input tokens. Actual savings vary by use case.
💡 Pro Tip: When building MCP servers, implement server-side TOON conversion. Your MCP server receives the API call (usually JSON from REST APIs), fetches the data, then converts the response to TOON before returning it to the LLM. The AI agent doesn't need to know about TOON—the conversion happens transparently on the server, and you get immediate 40-60% token savings.