Don’t Delete My Files: Securing the MCP Layer
Don’t Delete My Files: Securing the MCP Layer This isn't just another tech meetup—it's a reality check for the Cursor-and-chill era. We're moving from writing code to approving vibes. And while that's great for your velocity, it's a structural nightmare for your local machine and core infrastructure.
Everyone is shipping at the speed of thought. You're connecting AI agents to your terminal, local workspace, and private databases using Model Context Protocol (MCP) in minutes. The vibes are high, the PRs are merging, and your local tools feel like magic.
But here's the truth no one is talking about: Most vibe-coded tool setups are black boxes wrapped in hope.
While your prompts are generating local file access, they're also generating massive security blindspots. In the real world, vibes don't stop prompt injections. A strange payload hidden inside an email or document could easily trick your coding agent into deleting files or running malicious terminal commands right on your machine.
The industry has split into two groups:
The Fast: Shipping agentic features that'll be compromised by a basic indirect prompt injection by next Tuesday because they trust raw, unshielded tool connections.
The Secure: The people at this event learning to bridge the gap between "it works on my machine" and "this won't run
rm -rfon our local hard drive."
The Toolkit: Hardening the Vibe
We are turning AI House into an active security sandbox. This isn't a lecture—we're handing you the open-source weaponry to defend your agentic boundaries. We will intentionally spin up highly vulnerable, un-sandboxed MCP servers, give everyone the green light to hack them, and then deep-dive into:
Inline Gateway Architecture: How to sit directly in the traffic lane between your AI client and backend tools to intercept and sanitize inputs before they hit your system.
Sandbox Isolation: How to configure secure, open-source proxy configurations to spin up isolated container runtimes, trapping rogue agent behavior safely out of reach of your host machine.
Least-Privilege Enforcement: How to map per-project tool allowlists so your model never sees or executes risky, unapproved system commands.
The Rundown
06:00 PM | The Vibe Check: Drinks, networking, and a live look at the state of the tool exploit in 2026.
07:00 PM | The Anatomy of a Failed Vibe: How unprotected MCP connections are being bypassed in the wild right now to compromise local file systems.
07:45 PM | The Attack Phase: Live interactive session. Bring your laptop and try to force an unshielded agent to bypass its system prompt and leak local files.
08:30 PM | The Shield Phase: Setting up our secure gateway configurations, spinning up isolated container runtimes, and locking down the MCP layer.
09:15 PM | Late Night Lab & The Stress Test: Open floor. Bring your own vibe-coded workflows, throw adversarial payloads against our newly hardened gateways, and let's see if they survive the chaos.
The Perks
🍕 Food & Drinks: Covered by AI House. Come hungry, leave dangerous.
🤝 Community: Bangalore's AI builders, engineering architects, and founders all in one room.
🛠️ Hands-on: This is 100% execution over consumption. Bring your laptop with Python and Docker/Podman installed.
The window of innocence for AI agents is closing. The developers who stay ahead aren't the ones who prompt the fastest—they're the ones who can secure the output.
Stop vibing. Start securing. Spaces are strictly limited to keep the floor hands-on.
HiDevs is a Builder Intelligence Platform for the Agentic Era We help developers transform ideas into production-grade AI agents through personalized guidance, intelligent evaluation, continuous feedback, and contextual memory. By combining specialized AI agents with developer intelligence, HiDevs understands each builder's goals, tracks their progress, and provides the insights needed to accelerate their journey. Through 100+ community experiences, builder initiatives, and ecosystem collaborations, HiDevs has engaged 15,000+ developers and is building the infrastructure that powers the next generation of AI builders.
AIHouse is the community layer powering this ecosystem. Built for founders, researchers, engineers, students, and AI enthusiasts, AI House creates a space where people consistently learn, share, collaborate, and grow together. Through initiatives like 100 Days of AI House, community gatherings, founder conversations, knowledge-sharing sessions, and industry collaborations, AI House has built a thriving network of 15,000+ community members who are actively shaping the future of AI. More than a community, AI House is where meaningful relationships, opportunities, and ideas are created every day.
Together, HiDevs and AI House are building one of the fastest-growing AI ecosystems where ideas turn into products, builders become founders, and collaborations turn into impact.