ClaudeR RStudio MCP Server

What ClaudeR - RStudio MCP Server does

ClaudeR - RStudio MCP Server gives MCP-based agents access to a live R session running in RStudio. Agents can execute R code, inspect the environment, generate plots, read the active editor buffer, and apply or suggest code changes. The server works against the session where data, fitted models, packages, and other objects already exist, so it is suited to interactive analysis and iterative development.

Beyond direct coding, the toolset includes task lists, session history, project-source searching, checkpoint and restore operations, and tools for research workflows. These include reference verification through CrossRef, clean-room script probing, manuscript or file extraction, CSV annotation jobs, systematic-review screening reports, and RStudio Viewer content retrieval.

How it works

Install the ClaudeR R package, configure an MCP client, and start the server from RStudio with the ClaudeR addin. Calls are routed to the connected RStudio session. The active editor buffer is distinct from the saved file on disk: get_active_document reports unsaved changes, while file reads reflect disk contents. Editing tools save by default, or can stage changes in the buffer with save=false; suggest_edit is intended for changes that require user approval.

Long-running computations can use execute_r_async. The background process receives explicitly marshaled input objects and can return named output objects to the main session. It has no access to the main session environment beyond those inputs, and packages needed by the job must be loaded inside that process. Jobs can be checked or cancelled while normal R execution continues.

Multiple agents can share one session, while separate RStudio sessions can be discovered and selected with list_sessions and connect_session.

Setup and configuration

Install the R package from GitHub through devtools, load ClaudeR, then run install_clauder() for Claude Desktop or Cursor. For Claude Code CLI setup, use install_cli(tools = "claude"). Start the RStudio-side server with claudeAddin().

The project indicates R version 4.0 or newer. The server is designed for RStudio and also supports Positron. Client configuration is handled through the installation helpers or the relevant MCP client setup.

Tools and capabilities

• Execute R code synchronously, including code that produces plots.
• Inspect R version, package count, variables, session history, and the RStudio Viewer pane.
• Read files from disk, including structured extraction from DOCX and PDF files.
• Search R, R Markdown, and Quarto project source files with regular expressions.
• Modify or insert editor content, target a selected path, and request approval before applying edits.
• Create and update analysis task lists.
• Run, monitor, and cancel background R jobs with optional input and output marshaling.
• Create, list, and restore checkpoints of the R global environment.
• Run isolated script probes, verify DOI references through CrossRef, and generate screening reports.
• Load and annotate CSV rows with schema validation and resumable jobs.

Limitations and notes

Background jobs cannot marshal database connections or open file handles, and they must load their own packages. Auto-marshaled inputs are snapshots taken when the job starts, so later changes in the main session are not included. A focused RStudio document is required for active-buffer operations unless a target path is supplied where supported.

get_async_result waits before checking a job to reduce repeated polling. Annotation jobs process rows in separate fresh subprocesses or supported model processes, preserve the original CSV, and can be resumed. Editing operations that save to disk can leave the editor buffer and file state different depending on the chosen save behavior, so callers should use the returned save status when deciding whether to reread a file.