simonpcouch.com
he/him - writing statistical software at Posit, PBC (nÊe RStudio)đĨ
simonpcouch.com, @simonpcouch elsewhere
151 posts
3,252 followers
180 following
Prolific Poster
Conversation Starter
comment in response to
post
If you're a CC and R user and want to try this out, use `pak::pak("posit-dev/acquaint")` and then `claude mcp add -s "user" r-acquaint -- Rscript -e "acquaint::mcp_server()"`.
More on acquaint, an R package that implements the Model Context Protocol (MCP) in R: posit-dev.github.io/acquaint/
comment in response to
post
This is indeed a risk. We've configured a default set of tools that we feel are atomic and reduce risk, rather than some sort of general `run_r_code()` tool. In general, I'd recommend only using model providers that satisfy your privacy needs, i.e. that you're comfortable passing your data to.
comment in response to
post
asking the big questions
comment in response to
post
this is the way
comment in response to
post
I'd be interested to try that out! Do you use copilot in VS Code?
This might do the trick: docs.github.com/en/copilot/c...
...where the `command` and `args` could mirror the Claude Desktop setup in the acquaint homepage.
comment in response to
post
Please do file an issue if you struggled to get your config working! I would anticipate any of Windows/Linux/macOS should work, as well as R processes from Positron / RStudio / the command line.
comment in response to
post
I haven't! Do let me know if you give it a go.
comment in response to
post
Woops, thank you! Just corrected on GitHub.
comment in response to
post
Here's the full chat transcript, for those interested: claude.ai/share/7dc9e2...
Many thanks to @shikokuchuo.net and @winstonchang.bsky.social for helping bring this to life.
comment in response to
post
I think that totally makes sense. I often do the same!
comment in response to
post
ha, all is well!đ
comment in response to
post
You got it! www.simonpcouch.com/blog/2025-05...
I agree that the character/feel is quite different in a way that's difficult to quantify.
comment in response to
post
đđđ
comment in response to
post
...already building. This was enough of a hitch that I never really tried to push its limits. (2/2)
comment in response to
post
I think that makes sense.
I was really impressed with Qwen 2.5 Coder's outputs but disappointed with its interface. Given that it didn't seem to "see" system prompts, especially longer ones, and instead required them to be phrased as user turns, I couldn't just "plug it in" to things I was... (1/2)
comment in response to
post
Hoping to dig into smaller models more seriously here soon! I've been waiting on a `-coder` release of Qwen3 to give it a more serious go, but definitely curious.
comment in response to
post
Amazing! That's great to hear!
comment in response to
post
Yes! This works up to a rate limit; while running this eval, I eventually ran into rate limit errors and had to put a card down.
comment in response to
post
The release also includes several improvements to documentation from @joscelinrocha.com! Thank you!