Proposing a Native Middleware Architecture for Antigravity: How mcpv Solves Context Bloat & Latency

,

Hi Google AI Team & Developers,

I am a heavy user of Project Antigravity and a strong believer in its “Agent-First” philosophy. However, while building complex applications on Windows, I encountered significant infrastructure bottlenecks that hindered the “Vibe Coding” experience:

  1. Infinite Loading: 60s+ initialization time due to repeated file scanning.

  2. Context Bloat: Agents repeatedly requesting the full repomix context, freezing local MCP servers and wasting tokens.

  3. UI/Render Lag: Lack of proper GPU acceleration flags on Windows environments.

To solve this, I reverse-engineered the MCP pipeline and built MCP Vault (mcpv)—a smart middleware proxy that sits between Antigravity and the tools.

:link: Repo: https://github.com/thekeunpie-hash/mcpvault


:rocket: Technical Achievements

mcpv is not just a tool wrapper; it is an architectural fix for the “Context Crisis” in agentic IDEs.

  • Zero-Latency Startup (Lazy Loading):

    • Before: Antigravity scans all directories at boot → 60s delay.

    • After: mcpv initializes instantly (<0.1s) and acts as a router. It only connects to upstream servers (e.g., filesystem, neon) when the agent explicitly requests a tool.

  • The “Smart Valve” (Context Defense):

    • It intercepts JSON-RPC traffic to detect redundant context requests.

    • First Request: Allowed (Full Repomix context injected).

    Subsequent Requests: intercepted and blocked with a “Cached” signal (~20 tokens), saving 90% of token costs and preventing server timeouts.

  • Booster Injection (Windows Optimization):

    • Automatically injects hidden Chromium flags (--enable-gpu-rasterization, --enable-zero-copy) and bypasses Admin privilege issues via RunAsInvoker.
  • Smart Router (Flattening Tool Calls):

    • New Feature: It aggregates all tools from multiple upstream MCP servers into a single “Menu.”

    • The agent no longer needs to know which server ID (rube, context7) hosts a specific tool. It just calls run_tool(name="query"), and mcpv routes it automatically.


:handshake: Strategic Proposal: Let’s Integrate This Natively

I built mcpv because I needed Antigravity to be faster today. However, this middleware architecture belongs in the core of Antigravity, not as a third-party plugin.

The “Gateway Pattern” and “Smart Valve” logic I’ve implemented are essential for scaling agents beyond simple tasks without exploding token costs or latency.

I am proposing a strategic acquisition or acqui-hire discussion regarding the mcpv architecture.

I have solved the critical friction points for Windows users and optimized the context pipeline. I would love to bring this technology and insight directly into the Google Cloud / Antigravity team to make this the undisputed best AI IDE on the market.

mcpv pro dashboard

Best regards,

Jinho Noh, Developer of MCP Vault

1 Like

Hello @thekeunpie,

Thank you for your feedback. We appreciate you taking the time to share your thoughts with us, and we’ll be filing a feature request.
To help us prioritize this request effectively, any additional details you can provide regarding the impact this feature would have would be very helpful.