Feature Request: Add configurable timeout option for long-running R tasks

[lazymark2]

Hi @IMNMV,

First, thank you for creating ClaudeR - it's an amazing tool!

I'm encountering an issue where MCP connections timeout after ~2 minutes when running long R analysis tasks (e.g., CellChat single-cell analysis which takes 15-20 minutes).

Feature Request

Could you add a configurable timeout parameter to allow longer execution times?

Current Workaround

I have to create .R scripts and run them manually in RStudio, which loses the interactive benefit of ClaudeR.

Suggested Default

• Default timeout: 600 seconds (10 minutes)
• Allow user configuration: 60-3600 seconds (1 min to 1 hour)

I'm happy to test this feature or provide more examples. Let me know if you need any clarification!

[IMNMV]

Thanks for letting me know.

Is the RStudio connections dropping and unrecoverable? That is, does the agent try some analysis, it takes a while, it errors and outputs: "Error: RStudio addin is not running. Please start the Claude RStudio Connection addin in RStudio.", then the agent assumes the connection is closed? Or, is this another issue altogether?

If it's the former, can you let me know if the RStudio connectio truly drops, or if you can simply say 'continue', or 'it's back' without doing anything to see if the connection is back?

I ask because when I run heavy computational analyses, it takes R quite a bit of time to finish. During which the connection to the agent drops. However, once the computation in R finishes, if you tell the agent to continue, it is still able to connect without any intervention. This does mean you need to poke and prod it which does reduce some autonomy. If this is the issue I can make an update.

Let me know if your issue is different than the one I've outlined.

[IMNMV]

I took a stab at this and found out that the Anthropic has standard MCP timeouts at ~25 seconds that are baked into long-running calls. To remedy this, if a large calculation is performed that will take > 25 seconds, the model can now use the async tool where it will receive a job id and can ping it every ~10 seconds until it completes. It's a bit hackish, but it solves the time-out problem. I've tested it on 5 minute runs and it doesn't time out.