You will get a production-grade MCP server built for your team's Claude workflow

Project details
Most teams using Claude hit the same wall: every session starts from zero. Context is lost, behavior drifts, and the AI can't access your internal data. An MCP server fixes this by giving Claude persistent memory and custom tools that connect directly to your systems.
I build production-grade MCP servers — not wrappers or tutorials. My own MCP server (cognitive-memory) handles hybrid search across 7,000+ entries with sub-second latency, combining vector similarity, keyword matching, and graph traversal on PostgreSQL + pgvector.
What you get: a deployed, documented MCP server tailored to your workflow. Whether that's searchable knowledge bases, persistent conversation memory, or custom tool integration — scoped to your problem, built to last.
I build production-grade MCP servers — not wrappers or tutorials. My own MCP server (cognitive-memory) handles hybrid search across 7,000+ entries with sub-second latency, combining vector similarity, keyword matching, and graph traversal on PostgreSQL + pgvector.
What you get: a deployed, documented MCP server tailored to your workflow. Whether that's searchable knowledge bases, persistent conversation memory, or custom tool integration — scoped to your problem, built to last.
AI Algorithms
Large Language Model, Transformer ModelAI Applications
AI Chatbot, AI-Generated Code, Conversational AI, Natural Language Generation, Natural Language UnderstandingAI Development Language
PythonAI Models
ChatGPT, GPT-4What's included
| Service Tiers |
Starter
$800
|
Standard
$2,500
|
Advanced
$5,000
|
|---|---|---|---|
| Delivery Time | 5 days | 14 days | 30 days |
Number of Revisions | 1 | 2 | 3 |
AI Model Integration | |||
Batch Normalization | - | - | - |
Database Integration | - | ||
Detailed Code Comments | |||
Image Upscaling | - | - | - |
MLOps | - | - | - |
Model Deployment | - | ||
Model Documentation | - | ||
Model Monitoring | - | - | - |
Model Testing & Optimization | - | - | - |
Model Tuning | - | - | - |
Natural Language Processing | - | - | - |
NLP Tokenization | - | - | - |
Pre-Training | - | - | - |
Prompt Engineering | - | - | |
Setup File | - | - | - |
Source Code |
About Stefan
Claude Code Architect & MCP Server Developer | AI System Integration
Berlin, Germany - 2:42 pm local time
The scaling wall in AI adoption is rarely model access. It's context persistence, agent coordination, predictable behavior across sessions, and where each piece of the stack sits in relation to the others. Most teams discover this the expensive way — after the first two quick wins, when integration starts fighting back.
I build the architecture underneath, so Claude becomes load-bearing infrastructure instead of a side experiment.
What I've built
cognitive-memory — Persistent, searchable memory for Claude. An MCP server with hybrid search across 7,000+ entries, sub-second latency: vector similarity, keyword matching, and graph traversal fused via Reciprocal Rank Fusion into one ranked result set. PostgreSQL + pgvector.
tethr — Deterministic behavior control for Claude Code. Four hook types (SessionStart, UserPromptSubmit, PreToolUse, Stop) control what loads, what's allowed, and what's blocked — before the model responds. Keyword-routing context injection (6–10x token savings), PreToolUse gate for unsafe actions.
njord — Multi-agent orchestration for solo operators. Three specialized Claude agents coordinate a freelancing pipeline through typed artifact files. Cross-agent isolation, no shared state, resumable across sessions.
All three run in my own daily workflow. What you'd hire me to build is what I already rely on.
## Background
Founder of a first-to-market SaaS for professional poker players (2018–2024, solo operator — architecture, backend, frontend, operations). Since mid-2024, focused entirely on AI systems architecture. Native German, advanced English.
## What I don't do
Frontend beyond what a working system requires. Brand or visual design. Marketing automation. If your project needs those, you need a different hire or a complementary one — and I'll tell you early if the fit isn't there.
## Best fit
Teams that want Claude as load-bearing infrastructure, not a side experiment. Companies past the "we tried ChatGPT once" phase, sitting on Claude or OpenAI licenses, who notice that adoption has plateaued and want a second look before hiring in-house.
Steps for completing your project
After purchasing the project, send requirements so Stefan can start the project.
Delivery time starts when Stefan receives requirements from you.
Stefan works on your project following the steps below.
Revisions may occur after the delivery date.
Scope & architecture
I define the MCP server structure, tools, and data model based on your requirements. You review before I build.
Build & deploy
Server development, testing, and deployment to your infrastructure. Progress updates throughout.