EGC: Persistent Memory for AI Coding Tools via MCP Servers

Extended Global Context

Felipe Marzochi developed EGC (Extended Global Context) to eliminate the need for re-explaining project context across AI sessions. The system supports 13 different AI tools through a single installation process.

Why This Matters

Most AI coding sessions start from zero, forcing developers to manually re-explain architectural decisions and failed approaches every time they switch tools or restart a session. While the monetary cost is low—approximately $0.08 saved per 20 sessions—the primary technical friction is the interrupted developer flow and the inefficiency of spending ~1,500 tokens per session to rebuild context that should be persistent.

Key Insights

• Context efficiency: EGC reduces the token cost of rebuilding session context from ~1,500 tokens down to ~200 tokens (Marzochi, 2026).
• State Management Pattern: Implements a get_state/update_state cycle where project memory is stored in SQLite at ~/.egc/state/ rather than within the IDE.
• Security Layering: The egc-guardian server provides a validation layer to block shell injection and unauthorized writes to sensitive filesystem paths.

Working Examples

Installation and configuration of EGC across supported AI tools.

npm install -g @egchq/egc
egc install

Practical Applications

• …Use case: Cross-tool workflows (e.g., moving from Cursor to Claude Code) where project state like ‘use Redis, not in-memory’ is preserved automatically. …Pitfall: Manual context repetition leads to ‘session start fatigue’ and wasted token expenditure.

References:

• https://dev.to/fmarzochi/i-built-persistent-memory-for-ai-coding-tools-one-install-every-tool-no-more-re-explaining-context-45om
• https://github.com/Fmarzochi/EGC