Speaker
Description
R is an interactive environment. Working effectively with R means being able to interact with a live session to inspect objects, view and iterate on plots, access help, and step through running code in the debugger. Making LLMs effective in R therefore means more than giving them a way to execute code: it means exposing R’s interactive affordances in a form the model can use.
This talk introduces mcp-repl, an open-source front end to R that plugs into existing agents such as Claude and Codex. It provides a long-lived REPL with persistent state, in-band help, plot capture, support for debugger- and readline-based workflows, guardrails for large outputs, responsive handling of both interactive and long-running computations, and OS-level sandboxing. In practice, mcp-repl makes general-purpose coding agents much more capable at package development, Shiny debugging, and iterative analysis, especially in long-running autonomous tasks without a human in the loop.
Because mcp-repl plugs into existing agents, it also reflects the constraints of a plug-in architecture. To make that contrast concrete, this talk also looks at Posit Assistant as an example of what becomes possible when the interface is integrated into an IDE for data analysis and designed to keep the human in the loop.
Together, Posit Assistant and mcp-repl illustrate two ways to bring R’s interactive affordances to an LLM: one through an integrated experience designed for close human collaboration, the other through a plug-in front end designed for autonomous work in a private runtime.
Additional Material or Paper
Project repository for mcp-repl: https://github.com/t-kalinowski/mcp-repl.
If you used AI tools or services to support the preparation of this submission, please state the name and reason for using each of them.
I used ChatGPT to copyedit and polish my initial draft of the abstract. I reviewed, revised, and take full responsibility for the final submission.
| Keywords: Please list up to 5 keywords to help us find the right session for your contribution. | LLM, agents, interactive programming, mcp, sandbox |
|---|---|
| Virtual Option | This submission is for onsite presentation only |
| Video Recording | Video sharing is fine |
| The author(s) agree(s) to take responsibility and be accountable for the contents of the submission and is/are authorized to present it. | Confirm |