You will get a deep audit of your Claude Code or MCP setup with a clear fix plan
Rising Talent
Project details
Most Claude Code and MCP setups don't fail loudly — they drift. A tool call quietly returns the wrong thing, a subagent inherits the wrong model, context gets summarized away, and you only notice when the output degrades or the bill climbs.
I audit the setup you already have and tell you exactly what's wrong and what to do about it. You get a written report: concrete issues ranked by impact, the risks behind them, and quick wins you can ship this week — plus a prioritized fix plan. On the top tier, I implement the top fixes and hand you a clean PR.
Background: I build production Claude Code agent systems and custom MCP servers in Python and TypeScript — hooks, subagents, workflow orchestration. This is the setup I run every day, so the audit is grounded in what actually breaks, not theory.
No fluff, and no 400-line rewrites of files that needed three lines changed. You'll know where you stand and what to fix first.
I audit the setup you already have and tell you exactly what's wrong and what to do about it. You get a written report: concrete issues ranked by impact, the risks behind them, and quick wins you can ship this week — plus a prioritized fix plan. On the top tier, I implement the top fixes and hand you a clean PR.
Background: I build production Claude Code agent systems and custom MCP servers in Python and TypeScript — hooks, subagents, workflow orchestration. This is the setup I run every day, so the audit is grounded in what actually breaks, not theory.
No fluff, and no 400-line rewrites of files that needed three lines changed. You'll know where you stand and what to fix first.
AI Algorithms
Large Language ModelAI Applications
AI-Generated CodeAI Development Language
PythonAI Models
ChatGPT, GPT-4What's included $120
These options are included with the project scope.
$120
- Delivery Time 3 days
- Model Documentation
- Prompt Engineering
About Bogdan
Senior Claude Code & MCP Engineer | Clinical AI, RAG, Multi-Agent
Kyiv, Ukraine - 9:56 am local time
13 years as a clinical psychologist before AI engineering — rare combination. I build agent systems that pass both engineering review and clinical scrutiny, because I've sat on both sides.
— What I build —
• MCP servers exposing internal tools to Claude Desktop / Claude Code — auth, rate-limiting, error paths, prompt-injection guards. Not generic forks; tuned to each client's workflow.
• Multi-agent systems: dispatcher → 3-7 specialized subagents → quality-gate layer → handoff protocol before context-compaction. Persistent memory, not prompt-stuffing.
• Hook + sandboxing layers: pre-action guards, post-action telemetry, watchdog daemons. The boring guarantees that make AI systems reliable instead of theatrical.
• Healthcare / clinical AI: PHI deidentification with crypto envelope chains, supervisor-grade audit logs, clinical formulation modules grounded in CBT/Schema/MAPS-track frameworks. Where most AI engineers don't have clinical context to know what to refuse.
— Recent work (anonymized where NDA) —
• EU clinical SaaS extension: 3-role architecture (patient / therapist / supervisor), 11,309 production lines across 8 files, full code-review 10+ rounds.
• B2B research pipeline: 430 manually verified contacts across 12 markets in 2 weeks. Custom triangulation + Airtable delivery. over 2x cold-database response rate.
• Open Claude Code reference implementation: 20+ specialized agents, MCP integrations, memory layer, hook system, bridge architecture. Production-grade, used as my own daily ops for 12+ months.
— Stack —
Anthropic Claude (Fable 5 / Opus 4.8 / Sonnet 4.6 / Haiku 4.5), MCP (stdio + HTTP), LangGraph, RAG (ChromaDB / vector DBs), Python, TypeScript, Node.js, n8n.
— How I work —
First engagement = fixed-price milestone: scope doc (day 1-2), working system (day 5-10), production handoff (day 12-14). One revision. Daily git pushes — you audit progress async.
Bilingual EN/UA/RU.
— CTA —
Send your use case (workflow + pain point + stack). I reply within 24h on weekdays with yes/no + rough scope, or a referral if not my fit.
Steps for completing your project
After purchasing the project, send requirements so Bogdan can start the project.
Delivery time starts when Bogdan receives requirements from you.
Bogdan works on your project following the steps below.
Revisions may occur after the delivery date.
Kickoff & scope
You share read-only repo/config access and your pain points. I confirm scope and what the audit will cover.
Audit
I review your CLAUDE.md, MCP servers, hooks, subagents, context handling, and cost/performance — and find what's broken or risky.