EGC – Shared Memory for Every AI Coding Tool

TL;DR

TL;DR: EGC is an MCP server that gives every AI coding tool on your machine a shared, persistent brain—memory, skills, and decisions sync automatically across Cursor, Claude Code, Copilot, Aider, and 20+ other agents.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: egc.how ← verified reachable
• Source repository: github.com/Fmarzochi/EGC ← README read end-to-end
• License: Apache-2.0 (verified via GitHub API license.spdx_id)
• HN launch thread: news.ycombinator.com/item?id=48623049

What Is EGC?

EGC (Extended Global Context) is an MCP server that acts as a shared brain for every AI coding tool on your machine. Once installed, any decision made, context captured, or memory stored in one AI tool becomes instantly available in all the others.

The README puts it plainly:

“EGC isn’t another memory tool. It’s the intelligence layer that lets every AI work as if it has been on your project since day one, in Cursor, Copilot, Claude Code, Codex, Aider, and any terminal agent.”

Supported AI tools (20+): Cursor, Claude Code, Copilot, Codex, Aider, and any terminal-based agent. Supported models: Claude, GPT-4o, Gemini, DeepSeek, Mistral, Groq, Cohere, Vertex AI, and OpenRouter (Qwen3, Llama 4, and more).

Key features from the README:

• Persistent project memory stored in ~/.egc, encrypted with AES-256-GCM
• Guardian: validates every command before execution, blocks dangerous writes
• Token Crusher: compresses git logs, test output, install noise, and large JSON by up to 90%
• Prompt library: 63 agents, 230 skills, 77 commands, 111 rules included
• Memory is branch-aware and never committed to git

Setup Workflow

Step 1: Install EGC

npm install -g @egchq/egc && egc install

This installs the EGC CLI globally and runs the interactive setup that detects and configures all supported AI tools on your machine.

Step 2: Start Using It

There is no step two. Open any of your AI tools and speak naturally:

• “hi” or “let’s continue”
• “remember this decision”
• “what did we decide about auth?”

Sessions connect automatically. Memory loads without any manual sync. Every open tab, terminal, or AI tool already shares the same context.

Step 3: Access the Dashboard

A live dashboard showing agent activity, token usage, and cost savings spins up automatically after installation. To query token savings directly:

# Ask in any language
"How much did I save on tokens this week?"

Deeper Analysis

Why a Shared Brain Matters

Most developers use two or more AI coding tools simultaneously—a conversational agent in the browser and a code-agent in the terminal, or a different agent per task. The problem: every session starts from scratch. Context has to be rebuilt, decisions re-explained, and patterns re-learned.

EGC solves this by acting as the single source of truth for project context. Memory lives locally (not in the cloud), is encrypted at rest with AES-256-GCM, and is scoped to the current git branch so context does not bleed across feature branches.

Guardian: Safety Without Extra Steps

EGC’s Guardian runs as a background guardrail, not a separate tool to invoke. It validates every command before execution, blocks writes that look destructive, and detects prompt injection attempts. The README frames it as “an invisible safety net that keeps context lean, actions safe, and workflows autonomous.”

Token Crusher

AI coding tools generate large volumes of repetitive output—git logs, test runners, dependency installs, and verbose JSON responses. Token Crusher compresses this noise before it reaches the model, preserving errors and warnings while cutting token usage significantly. The 90% compression claim on git logs and install output is notable for long-running agent sessions.

Practical Evaluation Checklist

• Installs with one command (npm install -g @egchq/egc && egc install)
• Works with 20+ AI coding tools without per-tool configuration
• Memory persists across sessions and is encrypted locally (AES-256-GCM)
• Guardian validates commands before execution automatically
• Token Crusher reduces context noise by up to 90%
• Prompt library includes 63 agents, 230 skills, 77 commands, 111 rules
• Memory is branch-aware and never committed to git
• Supports Claude, GPT-4o, Gemini, DeepSeek, Mistral, Groq, and more
• OpenRouter support for Qwen3, Llama 4, and other open models
• Available in 12 languages (English, Arabic, German, Spanish, French, Hindi, Italian, Japanese, Korean, Portuguese, Russian, Turkish, Chinese)

Security Notes

• All memory stored locally in ~/.egc, encrypted with AES-256-GCM
• Memory is never committed to git (configurable per-branch awareness)
• Guardian provides command validation and prompt injection detection
• No cloud sync by default—context stays on your machine

FAQ

Q: Does EGC work offline? A: Yes. All memory is stored locally and encrypted on your machine. There is no required cloud service for core functionality.

Q: Which AI models are supported? A: Claude, GPT-4o, Gemini, DeepSeek, Mistral, Groq, Cohere, and Vertex AI natively. OpenRouter support covers Qwen3, Llama 4, and other open models.

Q: How is memory scoped between branches? A: EGC is branch-aware. Memory loaded in one git branch does not bleed into another, preventing context contamination across feature work.

Q: What happens if I use two different AI tools at the same time? A: Both tools share the same living context simultaneously. A decision made in one Cursor tab is known in a Claude Code terminal session instantly.

Q: Can I disable the prompt library and just use the memory feature? A: Yes. The README states that EGC’s persistent memory works exactly the same without any of the prompt library components.

Conclusion

EGC solves the context fragmentation that comes from using multiple AI coding tools. By acting as a shared, encrypted, persistent brain accessible to every agent on your machine, it eliminates the need to re-explain project context at the start of every session. The Guardian and Token Crusher features add safety and cost-efficiency on top of the core memory sync.

Installation is a single command. The 90% token compression on noisy output is a concrete efficiency win for long-running agent workflows.

Try it at egc.how or clone the repo at github.com/Fmarzochi/EGC.