Kette — Florentin One KI-Wertschöpfungskette

AI-First Enterprise Solutions for the German Market

@florentin-one/kette is a single, unified Model Context Protocol (MCP) server packaging all seven Florentin One reasoning tools into one Cloudflare Worker. Built for enterprise environments with GDPR compliance, German data sovereignty, and zero-downtime deployment.

Architecture

A single @florentin-one/kette package exposes all seven reasoning tools through one MCP endpoint. The server runs on Cloudflare Workers with Durable Objects for state management, deployed exclusively within EU data jurisdiction.

┌──────────────────────────────────────────────┐
│  @florentin-one/kette (single package)         │
│  ┌────────────────────────────────────────┐  │
│  │  MCP Server (Streamable HTTP)          │  │
│  │  metacognitiveMonitoring               │  │
│  │  sequentialthinking                    │  │
│  │  collaborativeReasoning                │  │
│  │  scientificMethod                      │  │
│  │  structuredArgumentation               │  │
│  │  constraintSolver                      │  │
│  │  narrativePlanner                      │  │
│  └────────────────────────────────────────┘  │
│  ┌──────────────┐  ┌──────────────────────┐  │
│  │  Code Mode   │  │  Agent SDK Handler   │  │
│  │  (direct TS) │  │  (createMcpHandler)  │  │
│  └──────────────┘  └──────────────────────┘  │
└──────────────────────────────────────────────┘

MCP Specification

Implements MCP 2026-07-28 using @modelcontextprotocol/server v2 with Streamable HTTP transport — a fully stateless protocol. Each request is self-contained; no session affinity or sticky routing required.

Installation

npm install @florentin-one/kette

A single package replaces the previous seven individual @florentin-one/kette-* packages.

Available Tools

| Tool | Description | | --- | --- | | metacognitiveMonitoring | Systematic self-monitoring: knowledge boundaries, confidence calibration, bias detection | | sequentialthinking | Dynamic step-by-step reasoning with revision, branching, and dependency tracking | | collaborativeReasoning | Multi-persona expert collaboration simulation with structured disagreement resolution | | scientificMethod | Formal hypothesis testing, variable identification, experiment design, evidence evaluation | | structuredArgumentation | Dialectical reasoning with thesis-antithesis-synthesis progression | | constraintSolver | Mathematical constraint satisfaction validation for numeric variables | | narrativePlanner | Three-act story structure planning with character development |

MCP Client Configuration

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "Florentin One Kette": {
      "command": "npx",
      "args": ["@florentin-one/kette@latest"]
    }
  }
}

Claude Desktop

Add to Claude Desktop configuration:

{
  "mcpServers": {
    "kette": {
      "command": "npx",
      "args": ["@florentin-one/kette@latest"]
    }
  }
}

Cloudflare MCP Portals

Connect via HTTP endpoint:

https://kette.florentin-one.de/mcp

Configure your MCP portal with the Streamable HTTP transport URL above. No additional tool-specific endpoints required — all seven tools are served from the single endpoint.

Code Mode API

Use reasoning tools directly in TypeScript without MCP transport overhead:

import {
  MetacognitiveCodeMode,
  SequentialThinking,
  CollaborativeReasoning,
  ScientificMethodCodeMode,
  StructuredArgumentation,
  ConstraintSolver,
  NarrativePlanner
} from "@florentin-one/kette";

const metacognitive = new MetacognitiveCodeMode();
const result = await metacognitive.monitor({
  task: "architecture review",
  stage: "knowledge-assessment",
  overallConfidence: 0.8,
  uncertaintyAreas: ["distributed consensus"],
  recommendedApproach: "systematic review",
  monitoringId: "mm-arch-20260820",
  iteration: 0,
  nextAssessmentNeeded: true
});

Cloudflare Deployment

A single Cloudflare Worker with Durable Objects for state management:

• Worker: kette — single entry point for all seven tools
• Durable Object: FlorentinOneMCP — SQLite-backed state persistence
• Jurisdiction: EU-only deployment, compliant with German data sovereignty requirements under GDPR Art. 28
• Transport: Streamable HTTP, stateless, no session affinity

Deploy from the package directory:

cd src/kette
pnpm exec wrangler deploy

Development

Prerequisites

• pnpm >= 11.20.0
• Node.js >= 22
• Git for version control

Setup

git clone https://github.com/florentin-one-cloud/kette.git
cd mcp
pnpm install
pnpm run build-all
pnpm run test:all

Project Structure

mcp/
├── src/
│   └── kette/                      # Kette — KI-Wertschöpfungskette MCP server package
│       ├── src/
│       │   ├── agent/              # MCP server factory (createServer)
│       │   ├── tools/              # Tool implementations
│       │   ├── codemode/           # Direct TypeScript API (7 tools)
│       │   ├── core/               # Pure business logic (7 tools)
│       │   ├── lib/                # Shared utilities (PostHog analytics)
│       │   ├── index.ts            # Stdio entry point + Code Mode exports
│       │   └── worker.ts           # Cloudflare Worker entry point
│       ├── package.json
│       ├── tsup.config.ts
│       ├── vitest.config.ts
│       └── wrangler.jsonc
├── .github/workflows/              # CI/CD pipelines
├── package.json                    # pnpm workspace root
└── pnpm-workspace.yaml

Dokumentation (HTAD-alignment)

Alle Dokumentation ist auf Deutsch und Englisch verfügbar, mit primärer Ausrichtung auf den DACH-regulierten Markt und die BMFTR-HTAD-Richtlinie "KI in Wertschöpfungsketten" (Juli 2026).

| Dokument | Inhalt | | --- | --- | | Grundlagen KI-Wertschöpfungskette | Positionierung, 7 Tools als Kettenglieder, vertikale vs. horizontale KI | | Strategisches Framework | HTAD-Alignment-Matrix, Technologische Souveränität, Blaupause-Transfer-Konzept | | Operative Methodik | LSTR 6-Stufen-Chain, MCP 2026-07-28 Transport, Cloudflare Deploy, GDPR Art. 28, Edge-AI | | Anwendungsfälle (DACH-Domänen) | Maschinenbau, Medizintechnik, Chemie/Cleantech, Agrar, IKT — je mit HTAD-Mapping | | Leitphilosophien | Augmentation über Automation, Systeme über Tools, Outcome über Output, Problem-First | | Formale Ethische Standards | EU AI Act (Art. 5 + Risikoklassen), WeMake Ethics, GDPR, HITL-Vier-Stufen-Modell |

License

MIT. See LICENSE for details.