GitHub - Fmarzochi/EGC: EGC gives every AI coding agent the same brain. Shared m

\n## One MCP Server to Unite All Your AI Coding Tools\n\nEGC (Everywhere Global Context) has a bold pitch: give Cursor, Claude Code, Copilot, Aider, and 20+ other AI coding tools a single shared brain. Developer fmarzochi posted a Show HN titled \"MCP server that gives AI coding tools memory across sessions.\" The project has 45 stars and 42 forks on GitHub. The HN post? 2 points, 0 comments. Cold start doesn't begin to describe it.\n\n## The Real Problem It Tries to Solve\n\nAnyone juggling multiple AI coding tools knows the friction. You spend twenty minutes explaining your project architecture to Cursor's AI, then switch to Claude Code in the terminal—and it has zero context. Every tool maintains its own conversation bubble. EGC attempts to fix this by running an MCP (Model Context Protocol) server that acts as a shared memory layer across all connected tools. Tell the AI something in one tool, and every other tool picks it up automatically. \"One brain. Everywhere\" is the tagline, and it's honest about the ambition.\n\n## How It Probably Works Under the Hood\n\nLooking at the repo structure, EGC is a standard MCP server implementation, likely built with Node/TypeScript. MCP is Anthropic's open protocol for connecting AI models to external tools and data sources. EGC essentially builds a cross-tool context synchronization layer on top of this standard. The \"zero configuration\" claim suggests it relies on MCP's auto-discovery mechanisms. The repo shows 3 branches and 45 tags, indicating active development—though the most recent commit is a dependency bump, not a feature addition.\n\n## The Honest Limitations Nobody Is Discussing\n\nWith 2 points and 0 comments on HN, the community hasn't exactly rushed to engage. There are likely reasons beyond mere discovery timing. First, native MCP support across tools is still spotty—the \"20+ tools\" number probably includes tools that support MCP only through indirect plugin chains. Second, \"shared memory\" glosses over serious engineering questions: context windows are finite, so how does EGC decide what to carry across sessions? Summarization? Retrieval-augmented lookup? The repo description is silent on this. Third, zero configuration sounds great until you think about permission scoping, data isolation between projects, and conflict resolution when two tools write contradictory context simultaneously.\n\n## Who Should Actually Pay Attention\n\nIf you're a power user running three or four AI coding tools simultaneously and constantly losing context between them, EGC points at a genuine pain point. But right now it's closer to a proof of concept than production infrastructure. The more interesting signal is the MCP ecosystem itself—as more tools adopt native MCP support, middleware like EGC becomes increasingly viable. For now, keep it on your radar, but don't slot it into your daily workflow yet.\n"===