MCP server for apps?
-
How about integrating MCP for apps? This way, I could use AI help, like Claude Code, for debugging or any other terminal access to any of my apps. It would need to be secured via SSH, but then it should be reasonably secure? And since this is only about app access, and not about full server access, there is little chance to break stuff on a bigger scale?
I suppose every app would need to have a separate MCP server, but I could just add the ones I really need.
-
-
Hello @ekevu123
I understand that the discussion is about OpenWebUI, but the technology discussedMCP
andMCPO
/MCP Proxy Server
are the same, or not?@ekevu123 said in MCP server for apps?:
I suppose every app would need to have a separate MCP server, but I could just add the ones I really need.
If I understand it correctly
MCP
should be added as a https://docs.cloudron.io/packaging/addons/ likeredis
for certain apps.Correct?
-
Isn't the better place for a mcp server in Cloudron directly at the Cloudron Api? Via mcp your ai could talk to the Cloudron Api, and through this access files, logs, execute commands, install/update.
The api is described here: https://docs.cloudron.io/api.html
I would think since there is an openapi spec there is probably already some mcp server that could import it.
-
Using the Cloudron API is a good idea to solve this! That would enable MCP capabilities, I suppose. But I think the feature request still stands here, I understand I could code a workaround right now, but it could be convenient to activate the MCP where needed, perhaps by copy-pasting a snippet, which should be fine.
-
Isn't the better place for a mcp server in Cloudron directly at the Cloudron Api? Via mcp your ai could talk to the Cloudron Api, and through this access files, logs, execute commands, install/update.
The api is described here: https://docs.cloudron.io/api.html
I would think since there is an openapi spec there is probably already some mcp server that could import it.
@fbartels An MCP server for Cloudron would be useful. I would like to be able to use Claude for orchestration and not (necessarily) need to rely on the cloudron npm cli tool.
However it would only be for Cloudron itself. Each app would still need to have an MCP written for it.
I am about to be spending all day every day in Claude and I am living the MCP life (currently Redmine/Joplin). And working on packaging (via Claude) about 50 apps for Cloudron.
-
Using the Cloudron API is a good idea to solve this! That would enable MCP capabilities, I suppose. But I think the feature request still stands here, I understand I could code a workaround right now, but it could be convenient to activate the MCP where needed, perhaps by copy-pasting a snippet, which should be fine.
@ekevu123 MCP is highly application specific. It is out of scope for Cloudron to have an "MCP server" (except, as discussed in the thread, perhaps for Cloudron itself).
Each app has it's own functionality to be exposed/consumed/orchestrated by an MCP client.
What use cases do you have in mind ? Can you describe how you would use MCP with Cloudron apps?
-
I wrote a toy MCP for Cloudron CLI , mostly to understand what the hell MCP is. https://git.cloudron.io/playground/toy-mcp-server/ . This already is able to list/start/stop/restart/show logs of apps! It's quite insanely powerful.
Here's a screencast if anyone cares
My code configuration is:
"mcp": { "servers": { "cloudron-cli": { "type": "stdio", "command": "node", "args": [ "mcp-server.js" ], "cwd": "/home/girish/yellowtent/tmp/toy-mcp-server" } } }
-
I wrote a toy MCP for Cloudron CLI , mostly to understand what the hell MCP is. https://git.cloudron.io/playground/toy-mcp-server/ . This already is able to list/start/stop/restart/show logs of apps! It's quite insanely powerful.
Here's a screencast if anyone cares
My code configuration is:
"mcp": { "servers": { "cloudron-cli": { "type": "stdio", "command": "node", "args": [ "mcp-server.js" ], "cwd": "/home/girish/yellowtent/tmp/toy-mcp-server" } } }
That is awesome Girish!
Definitely not a toy.
Glad you're able to use AI tools now.
For another level of code analysis, check out the newly released https://github.com/google-gemini/gemini-cli/
It let's you examine the entire code base with a huge context window, and look for say ... obvious bugs, or add whole new features. Can use Cursor as the editor.
It's also free to use if you auth with a personal gmail account; has daily limits.
Edu accounts can get a year free, and API has full access.
-
I wrote a toy MCP for Cloudron CLI , mostly to understand what the hell MCP is. https://git.cloudron.io/playground/toy-mcp-server/ . This already is able to list/start/stop/restart/show logs of apps! It's quite insanely powerful.
Here's a screencast if anyone cares
My code configuration is:
"mcp": { "servers": { "cloudron-cli": { "type": "stdio", "command": "node", "args": [ "mcp-server.js" ], "cwd": "/home/girish/yellowtent/tmp/toy-mcp-server" } } }
-
I am not so sure if every app would need a specific MCP server written for it, because as I see it, the MCP server should expose only the cloudron-specific functionality of any app, not allow with the app itself.
What that means specifically: Using Claude Code, Cursor etc. I could prompt "check the production logs of my app to investigate why the hell this doesn't work".
Ideally, via MCP, the AI could execute test scripts that have been uploaded before via terminal, or do similar basic things such as go through files of the app.I would be more interested in app-based interactions rather than the whole cloudron itself, but that use case might differ between people.