Kongen MCP — Pattern Intelligence
kongen-labs/kongen-mcp
Detecção de complexidade e roteamento de modelos Claude via Pattern Intelligence
Ver no GitHubServidor MCP que conecta Claude, Cursor e Windsurf à API de Pattern Intelligence do Kongen Labs. Oferece detecção de regime de raciocínio, scoring de transferência estrutural e roteamento inteligente entre modelos Claude com base em complexidade de prompts.
Instalação
npx convez-ai kongen-labs/kongen-mcp1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Analisar complexidade de prompts e alocar token budget ótimo
- Recomendar modelo Claude (Haiku/Sonnet/Opus) baseado na tarefa
- Avaliar similaridade estrutural entre padrões cross-domain
- Monitorar uso de tokens da API Kongen
Veredito

Ferramenta niche mas bem integrada para quem usa sistematicamente a stack Claude + Kongen. Adiciona inteligência na alocação de modelos, evitando overspend com Opus em tarefas simples — útil em cenários de produção com múltiplas requisições.
README
A tradução para o português deste README ainda está na fila.
kongen-mcp
MCP (Model Context Protocol) server for the Kongen Labs Pattern Intelligence API.
Connects Claude Code, Cursor, Windsurf, and any MCP-compatible client to Kongen's cross-domain pattern intelligence — reasoning regime detection, structural transfer scoring, and intelligent model routing.
Tools
| Tool | Description | Cost |
|---|---|---|
score_prompt | Detect reasoning regime and optimal token budget for a prompt | 1 KT |
transfer_score | Score a structural signature against cross-domain reference patterns | 50 KT |
check_usage | Check your token balance and usage | 0 KT |
route_model | Recommend Haiku/Sonnet/Opus based on prompt complexity | 1 KT |
Installation
pip install kongenlabs-mcp
Or install from source:
git clone https://github.com/kongen-labs/kongen-mcp.git
cd kongen-mcp
pip install .
Configuration
Get your API key at kongenlabs.life.
Claude Code
Add with the CLI:
claude mcp add kongen -- kongen-mcp
Then set your API key when prompted, or add it manually to .claude/mcp.json:
{
"mcpServers": {
"kongen": {
"command": "kongen-mcp",
"env": {
"KONGEN_API_KEY": "kl_live_..."
}
}
}
}
Cursor
- Open Settings > MCP > Add Server
- Name:
kongen - Command:
kongen-mcp - Environment variables:
KONGEN_API_KEY: your API key
Windsurf
Add to your MCP configuration:
{
"mcpServers": {
"kongen": {
"command": "kongen-mcp",
"env": {
"KONGEN_API_KEY": "kl_live_..."
}
}
}
}
Custom base URL
To point the server at a different API endpoint (for example a local proxy), set KONGEN_API_BASE_URL:
{
"mcpServers": {
"kongen": {
"command": "kongen-mcp",
"env": {
"KONGEN_API_KEY": "kl_test_...",
"KONGEN_API_BASE_URL": "http://localhost:8000"
}
}
}
}
Usage examples
Once configured, the tools are available to your AI assistant. Examples of what you can ask:
Prompt scoring:
"Score this prompt for complexity: Prove that the square root of 2 is irrational using proof by contradiction."
Model routing:
"Which Claude model should I use for this task: Summarize this 3-line email."
Transfer scoring:
"Score this structural signature against cross-domain patterns."
Usage check:
"How many Kongen tokens do I have left?"
Environment variables
| Variable | Required | Default | Description |
|---|---|---|---|
KONGEN_API_KEY | Yes | — | Your Kongen API key (kl_live_... or kl_test_...) |
KONGEN_API_BASE_URL | No | https://api.kongenlabs.life | API base URL override |
License
MIT