We shipped a streaming MCP server - connect Cursor, Windsurf, or Claude Desktop with one URL
Cloud World Model now has a streaming MCP server. Paste the URL into Cursor, Windsurf, or Claude Desktop and simulate AWS/GCP/Azure/OCI/DigitalOcean architectures, inject chaos, run RL training - no install, no API key required to start.
What we shipped: a hosted Streamable HTTP MCP server for cloud simulation. You paste one URL into your AI client, and it can simulate production infrastructure, inject failures, analyze costs, and run RL training episodes - all without provisioning anything real.
The zero-friction angle is the lead story. No npx install, no Docker container, no local setup. Just:
https://www.cloudworldmodel.ai/mcpDrop that into @Cursor (mcpServers config), @Windsurf , or @Claude for Desktop remote servers, and you're done. We also listed it on Smithery for one-click connection.
Two tiers:
No API key - 6 demo tools, 50 total calls, 20 sim steps. Enough to run a real simulation and see what the server does before committing to anything.
API key - all 49 tools, no caps. Architecture validation, chaos testing, multi-cloud cost comparison, RL agent training, AI-generated recommendations.
Who it's for: AI engineers who want to test cloud architecture decisions from inside their coding environment without spinning up real infrastructure. Cursor users who want to ask "will this design handle 10k RPS?" and get a real answer. RL researchers who need a fast, free simulation environment for cloud optimization policies.
We built the simulation engine first (97%+ accuracy against real AWS/GCP/Azure/OCI benchmarks). The MCP server is the interface layer that makes it accessible from any AI toolchain.
Would love to hear what you build with it - or what's missing from the current tool set.
Replies
If you looked at your usage analytics today, whatss the one metric that tells you users are actually getting value instead of just experimenting?
Cloud World Model
@shiv_kumar_sharma4 Right now it’s early, but one signal I’m watching closely is whether users go beyond creating a simulation and continue stepping it, injecting traffic or failures, and iterating. That tells us they’re using it to explore decisions, not just trying the demo once. As usage grows, we’ll be watching repeat simulations and authenticated usage even more closely. It's a little bit of a new concept and it's e early.
Also, we've implemented PostHogs new MCP analytics which is pretty cool thus far. It shows the amount of users, sessions, tool calls, etc. It's in beta but if anyone else implemented a mcp server it's worthwhile trying out. https://posthog.com/docs/mcp-analytics/installation
Love the direction. How do you balance simulation speed with accuracy? At what point do you decide a slower but more realistic simulation is worth the tradeoff?
Cloud World Model
@joseph_parker3 Great question. We use a hybrid approach: a fast rule-based simulator backed by an ML layer that captures provider-specific behavior like autoscaling and latency patterns. For exploration, speed is critical. For validation, we’re happy to spend a little more time if it helps avoid an expensive architectural mistake in production.
The multi-cloud stuff is cool as well. One cool question to try asking the simulator is "When is OCI cheaper than Digital Ocean?". One might be surprised at the answer.
Really interesting launch. Have any users already found edge cases where the simulation behaved differently than the real cloud provider? Those stories are usually fascinating.
Cloud World Model
@ella_reyes1 Thanks! Yes and those are actually our favorite bug reports. One example: we found GCP’s autoscaler behaves differently than AWS at moderate CPU loads, so we updated the simulator to match real-world behavior. We’re continually comparing simulations against real cloud behavior because those edge cases are where the model becomes most valuable.
Buffup.AI
Five years from now, do you imagine engineers asking AI before every infrastructure change, with Cloud World Model running silently in the background? Or is the bigger vision something completely different?
Cloud World Model
@sansa_grey I think that’s where we’re heading. The MCP server is an early step, but the bigger vision is AI agents continuously simulating infrastructure before changes are made predicting cost spikes, bottlenecks, and resilience issues before they become incidents. We’d love for “I tested it in Cloud World Model first” to become as standard as “I wrote a test first.” The goal isn’t replacing engineers, it’s giving them confidence before production. The way it is today. One actually has to provision resources or rely on best practices for architectural confidence. It's a little backwards.
Really interesting direction. Im curious what's the biggest misconception people have after reading the landing page? Is there one capability everyone assumes the MCP server has (or doesn't have) that you've had to explain repeatedly since launch?
Cloud World Model
@jordan_bulk The biggest misconception is that Cloud World Model is a cloud provisioning tool. It isn’t. It simulates cloud infrastructure so AI agents and engineers can test ideas before spending money or touching production. The MCP server exposes that simulation engine, so agents can reason about infrastructure without needing real cloud resources.