Running DOOM in ChatGPT and Claude: Exploring the Frontiers of Model Context Protocol (MCP) Apps

The classic query, "Can it run Doom?" has long served as a litmus test for new computing platforms, from smartwatches to pregnancy tests. This enduring challenge has now found its way into the realm of large language models (LLMs) and AI chatbots, with an intriguing experiment demonstrating DOOM running as a Model Context Protocol (MCP) app within environments like ChatGPT and Claude. While the original blog post detailing this specific implementation became temporarily unavailable due to usage limits, the community discussion around it illuminates the profound implications and capabilities of MCP apps.

This endeavor isn't about the AI itself running DOOM in a computational sense. As one commenter clarified, "The title is misleading. This is in ChatGPT but not on ChatGPT. It's not running on AI. It's running on a site that serves an AI chatbot." Instead, it leverages MCP to render web resources—essentially HTML in an iframe—directly within the chat interface, transforming the chatbot into a display harness for interactive applications.

Understanding Model Context Protocol (MCP) Apps

At its core, the Model Context Protocol (MCP) enables AI models to interact with external tools and services. While often perceived as a bridge to standard utilities like Google Drive, its true potential for powering "actual in-chat apps is very under-explored," as noted by @firasd. MCP allows developers to register resources that, when called by the AI, render content directly within the chat window. This content is typically HTML, displayed in an iframe, providing a rich, interactive experience that goes far beyond simple text responses.

"Cool, but yes you can render whatever you want as long as you register an MCP resource and use it as part of a tool call. It is just html in an iframe." — @pdp

This mechanism fundamentally shifts how we perceive AI chatbots, moving them from mere conversational interfaces to potential platforms for a new generation of chat-native applications.

DOOM as a Benchmark for Emerging Tech

The choice of DOOM for such an experiment is no accident. Its historical role as a performance benchmark for new hardware makes it an ideal candidate for stress-testing emerging software paradigms like MCP apps.

"Snake and DOOM were two of our early tests (for filter functions and MCP) when we stood up Open WebUI for internal chat/agent use. Sometimes games are the best way to limit-test new tech." — @alach11

This sentiment underscores the value of games in pushing the boundaries of new technology, revealing limitations and inspiring innovative solutions. The question "Can it run Doom?" evolves with technology:

"I wonder what the future 'can it run doom ?' Will it be: Can it run AGI on a new quantum processor/ robot / bio-engineered cell. Or will it be Doom all the way down..." — @phodo

Beyond Games: The Broader Potential of In-Chat Applications

The ability to embed interactive web content opens a vast array of possibilities beyond just gaming. Commenters highlighted several practical and innovative use cases:

• Data Surfacing: Imagine interactive tables or dynamic data visualizations directly within your chat. @firasd demonstrated this with "Liveclip," a project that allows ChatGPT to surface sentences within the US Constitution, acting like "tables in chat."
• Utilities: Simple, yet effective tools like a clock or a calculator could be integrated directly into the chat experience, removing the need to switch contexts.
• Educational Tools: Interactive tutorials or simulations could be embedded, enhancing learning within a conversational flow.
• Productivity Apps: Small, focused applications that streamline workflows without leaving the AI interface.

@firasd even shared a no-auth remote MCP clock, demonstrating the ease of deployment and accessibility of such tools: "In fact I have a no-auth remote MCP clock that can be added to any client right now deployed straight from https://github.com/firasd/mcpclock to https://mcpclock.firasd.workers.dev/mcp <-- works in Claude .ai, Claude iOS, Claude Code, Codex, etc right now."

Technical Considerations and Challenges

Implementing MCP apps, especially those involving rich media, is not without its technical hurdles. Content Security Policy (CSP) is a significant concern, particularly when serving static resources like images or game assets.

"I'd love if you could explain the issues you had with CSP and serving content. I was just tinkering around with MCP Apps the other day, and I had a really hard time serving images (this is for rendering in VSCode Copilot). Did you serve static resources through resource endpoints in the mcp server, or via another mechanism?" — @JambalayaJimbo

These challenges highlight the need for careful architecture and configuration to ensure that embedded content can load and function correctly within the secure environment of the chat harness. Solutions often involve serving static resources through dedicated endpoints or carefully managed Content Delivery Networks (CDNs) with appropriate CSP headers.

A Community of Innovators

The idea of embedding rich, interactive content within AI interfaces is not entirely new, and the community has been actively exploring similar concepts. Prior to this DOOM demonstration, other developers had already pushed the boundaries:

• Bad Apple: @eigenblake successfully embedded the iconic "Bad Apple" video into an MCP App, showcasing its multimedia capabilities.
• Earlier DOOM Implementations: It was noted that similar DOOM implementations had been demonstrated previously, indicating a shared curiosity within the developer community.
• Snake: Alongside DOOM, the classic game Snake was also used in early tests for MCP and filter functions, further cementing games as valuable testbeds.

These parallel efforts underscore a collective recognition of MCP's potential and a collaborative spirit in exploring its frontiers.

Evolving Perspectives

Ultimately, projects like running DOOM in an MCP app serve as powerful demonstrations that can fundamentally alter our understanding of what AI chatbots are capable of. They move the conversation beyond text generation to interactive experiences, hinting at a future where AI interfaces are not just conversational partners but also platforms for a new class of applications.

"Apart from a cool project, this evolved my perspective on what an MCP is, along with some cool architecture insights and inspiring ideas. Thank you!" — @neosat

The experiment, despite the original site's temporary unavailability, has successfully sparked a dialogue about the untapped potential of MCP, inspiring developers to think creatively about how AI chatbots can host dynamic, interactive experiences.