The silent tax every developer pays
Every session starts from zero. The AI does not know what you decided last week. It does not know what failed. It does not know your stack preferences or the architectural direction you chose three sessions ago.
Rebuilding that context costs time and tokens -- around 1,500 tokens per session. Multiply by 20 sessions a month and you have lost over 3 hours just re-explaining yourself.
No company has solved this. Not Anthropic, not OpenAI, not Google. The AI always resets.
I got tired of it and built EGC -- Extended Global Context.
The solution: two local MCP servers
egc-memory
Persistent state across sessions. At the start of every session, the AI calls get_state({}). It reads a plain Markdown file from ~/.egc/state/ and knows exactly where you left off.
At the end, it calls update_state({...}) with decisions made, what to avoid, and what to pick up next.
egc-guardian
Command validation before execution. Blocks shell injection, blocks writes to sensitive paths, enforces trust levels. Runs silently in the background.
One install, every tool
npm install -g @fmarzochi/egc && egc install
Detects Claude Code, Cursor, Codex, Gemini CLI, Kiro, OpenCode, Trae, CodeBuddy and registers both MCP servers in all of them.
Token savings
~200 tokens per session vs ~1,500 rebuilding from scratch. No cloud. No subscription. Free and open-source.
Try it
- GitHub: https://github.com/Fmarzochi/EGC
- Site: https://fmarzochi.github.io/EGCSite/
- npm:
@fmarzochi/egcWould love feedback from anyone using multiple AI tools.
Top comments (0)