After seeing the release video for Unreal Engine 5.8 and checking out the new AI/MCP plugins, I just couldn’t wait to get my hands on it and test this thing out. Wanted to share my experience so far, because honestly, we’re dealing with some super early experimental stuff here, but the potential is huge.
Here is a breakdown of what I’ve been up to today trying to bridge UE 5.8 with local and default LLMs
I’m far from an expert in AI and that whole field, so all of this was new to me, and I was learning as I went along.
First thing I did was go to the plugins section and find these two plugins:
As I understand it, UNREAL MCP collects data within the engine that can be passed to the AI
The MCP Client Toolset is some kind of client that actually connects to the AI
I restarted the engine and, of course, started looking for new plugins. I looked absolutely everywhere but couldn’t find them anywhere the only thing I found were some commands in the output log.
All commands begin with “ModelContextProtocol”
I started the server using the “ModelContextProtocol.StartServer” command. After running this command, the engine displays this text and successfully starts the server “LogModelContextProtocol: Starting MCP server on port 8000 (override with -ModelContextProtocolPort=N).
LogModelContextProtocol: Warning: Data transmitted via this plugin to your connected LLM service is Licensed Technology under the UE EULA. You are responsible for ensuring your LLM provider does not use it as training input. See Section 6(e) of the UE EULA for full terms.
LogHttpServerModule: Starting all listeners…
LogHttpServerModule: All listeners started”. Everything is working fine rn
you can use the “ModelContextProtocol.GenerateClientConfig” to generate the configuration for connecting to the AI
The engine generates configurations for the neural networks listed above
After this command, the engine adds these files to your project folder
For some reason, the configuration file for Claude is located outside the folder
The Roadblocks (Where things got messy)
Honestly, this is where I hit a massive brick wall, and it took me hours of troubleshooting. I tried absolutely everything to get a connection, switching between different clients and AI tools, but kept getting hit with errors left and right.
Here is a quick list of what I attempted:
Claude Desktop: Since the engine generated a config for Claude, I thought it would be a plug-and-play situation. Nope. Claude threw an error on startup saying the configuration wasn’t valid. It seems Epic’s plugin uses a raw HTTP/SSE format (“type”: “http”), which Claude Desktop doesn’t natively accept yet (To be honest, I don’t fully understand what that means myself that’s what the AI told me xD )
Cursor: I tried adding the server manually through Cursor’s MCP settings using the HTTP type. At first, it just kept saying fetch failed.
VS Code + Roo Code: I didn’t give up and moved to VS Code. I even tried setting up an internal proxy command (npx @modelcontextprotocol/server-http-sse) to bridge the HTTP stream into something the extension could read. It actually managed to ping the engine for a second, but then UE instantly killed the socket with a Connection closed (Error -32000)
**
Conclusion**
Even though I couldn’t get my local models or Claude to actually read my project data yet, it was a wild ride.It seems the built in HTTP server is just too unstable in this early preview and drops connections instantly. We are literally on the bleeding edge here. If anyone has figured out how to keep the connection stable or knows a specific proxy setup that UE 5.8 actually likes, please let me know!