Now in founding member preview

AI gets dumber the more you feed it. We engineeredwhat it sees.

Huri cognitively cleans MCP tool responses — 70-90% noise removed. One word in your config. Every MCP server, instantly sharper.

~/.config/claude/claude_desktop_config.json
// One word. That's the entire install.
{
"mcpServers": {
"notion": {
"command": "huri",
"args": ["wrap", "--", "npx", "notion-mcp"]
}
}
}

How Huri works

Four cognitive cleaning layers. Each one independent. Combined: 70-90% noise removed.

Notion API Response13-block page query
Before Huri5,290 tokens
After Huri146 tokens
Reduction36x smaller
1

Smart Field Stripping

5-10x reduction

Domain-aware deny-lists strip metadata bloat. Notion's all-false annotations, GitHub's avatar URLs, null fields, empty arrays — gone.

Before
"bold": false, "italic": false,
"strikethrough": false,
"underline": false, "code": false,
"color": "default"
After
// stripped (all defaults)
2

Tabular Encoding

2-3x reduction

Arrays of similar objects become header + rows. 50 Notion blocks with 12 keys each: 600 key repetitions become 1 header row.

Before
[{"type":"paragraph","id":"a1"},
 {"type":"heading","id":"a2"},
 {"type":"paragraph","id":"a3"}]
After
type,id
paragraph,a1
heading,a2
paragraph,a3
3

ID Side-Table

context-critical reduction

UUIDs replaced with short refs. The proxy holds the mapping and re-expands on write-back. Lossless round-trip.

Before
"id": "380c78c0-e0f5-4a8b-
       9e3c-7d2f1a..."
After
"id": "b1"
4

Tool Definition Compression

5x reduction

On initialize, compress tool schemas themselves. Strip verbose descriptions, collapse enums, deduplicate shared parameters.

Before
75,000 tokens of tool
definitions (10 servers)
After
~15,000 tokens
0%
Average noise removed
Across Notion, GitHub, Slack
0x
Peak compression ratio
Notion API responses
0
Cognitive cleaning layers
Each independent, all composable
Built on Rust + MCP protocol
Zero data leaves your machine
Offline license — no phone-home

Simple pricing

Free compression core. Pay for translation and tools. Every tier pays for itself in saved tokens.

Free

Start compressing. Zero risk.

$0
  • All 4 cognitive cleaning layers
  • Unlimited MCP servers
  • Basic stats
  • .ember output
  • Advanced dashboard
  • Model scoring
Recommended

Pro

Compression + translation.

$19/mo
$12/mo founding
  • All 4 cognitive cleaning layers
  • Unlimited MCP servers
  • Basic stats
  • .ember output
  • Advanced dashboard
  • Model scoring

All prices USD. No hidden fees. Cancel anytime. Huri runs locally — your data never touches our servers. Offline license validation. No phone-home.