I build automation workflows, MCP servers, agent infrastructure, observability tools, and local desktop applications.
📊
01 / TOKEN OBSERVABILITY— tokenscopeLocal-first usage and subscription tracker for Claude Code and Codex. Costs, quotas and budgets — without telemetry.
PythonReactFastAPI
📈
02 / BUSINESS DATA BRIDGE— powerbi-mcp-localMCP server for Power BI Desktop: Excel, Power Query, semantic models and visual automation.
PythonMCPPower BI
🛰️
03 / DESKTOP SIGNAL— claude-rpcNative Discord Rich Presence for Claude Code and Desktop, with tray controls and usage limits.
RustTauriRPC
🔌 RACK A — AGENTS CONNECTED TO REAL SYSTEMS
| Project | Signal |
|---|---|
| obsidian-mcp | MCP tools for reading, searching, and writing an Obsidian vault. |
| steam-mcp | Steam profiles, libraries, achievements and store search over MCP. |
| mcp-worldoftanks | Wargaming.net account analytics for Claude and Gemini. |
| gemini-mcp-connect | Gemini review tools exposed inside Claude Code through MCP. |
| codex-discord-mcp | Discord bot and MCP bridge for the Codex CLI. |
🧠 RACK B — AGENT OPS, MEMORY & GUARDRAILS
| Project | Signal |
|---|---|
| agentops-config-toolkit | Repository-level scope, context, and output rules for coding agents. |
| remnant | Local-first persistent context for Claude Code, Codex and Gemini CLI. |
| security-hardening | Defensive AppSec and AI-agent security skill. |
| caveman-lang | Compact-output instruction protocol for agent responses. |
| prompting-evaluator | Compares prompt behavior with local or hosted models. |
| agentic-coding-FR | Hands-on French guide to Claude Code and Codex. |
| ai-edu-skills-FR | ChatGPT and Claude skills for French academic writing. |
🧪 RACK C — DESKTOP TOOLS, LIVE LABS & FIELD TESTS
| Project | Signal |
|---|---|
| codex-rpc | Native TypeScript/Tauri/Rust Rich Presence for Codex. |
| icue-edge-widgets | Spotify, ISS and dashboard widgets for Corsair XENEON EDGE. |
| Deadzone Lab | Test and tune PlayStation controller deadzones on PC. |
| Nihongo Lab | Japanese learning from kana to JLPT N1, with SRS. FR/EN. |
| osint-terrain | Provenance-first retrospective terrain analysis. |
| Cruise | Forza EventLab automation with telemetry and stuck recovery. |
| pitwall | Forza telemetry overlay for OBS and transparent desktop HUDs. |
LANGUAGES + DATA
PythonTypeScriptRustSQLPowerShellVBAPower BIExcel
AGENTS + CONTROL
ClaudeAnthropic SDKCodexGeminiCursorMCPPrompt evalsAgent memoryAppSec
SHIP + OBSERVE
TauriReactFastAPIAcceptance testsAgile/ScrumDocumentationChange management
LOG 01 — The agents write the code. I write the rules they break, then the rules that stop them breaking them. The second file is longer.
LOG 02 — The tools have better observability than most production systems I've seen. This is either reassuring or deeply concerning.
LOG 03 — Claude Code, Codex and Antigravity don't know about each other. gemini-mcp-connect exists so they can argue.
🖥️ WORKSTATION SPEC — THE BOX RUNNING THE AGENTS
| Part | Spec |
|---|---|
| CPU / GPU | AMD Ryzen 7 9800X3D · Gigabyte RTX 5070 Ti AERO 16GB |
| Memory | HyperX FURY Beast RGB 64GB DDR5-6400 CL32 |
| Board / cooling | GIGABYTE AORUS B650E STEALTH ICE · Corsair iCUE Link Titan 360 RX |
| Storage | 2× WD SN7100 2TB · Samsung 970 EVO Plus 2TB · WD SA500 4TB |
| Case / PSU | Corsair 4000D Frame White · ASUS ROG Strix 1000W Platinum |