IMNMV/ClaudeR

devtools::install_github("IMNMV/ClaudeR")
library(ClaudeR); install_clauder()   (Claude Desktop / Cursor)
install_cli(tools = "claude" | "codex" | "qwen" | "agy" | "gemini")   (CLI agents)
PyPI bridge: uvx clauder-mcp

Usage

• •library(ClaudeR); claudeAddin() in RStudio, then click Start Server
• •Ask the agent to run analysis/code in your active RStudio session; plots are captured and visible to the model
• •reviewer_zero_prompt() for a 4-pass (up to 6-pass) automated manuscript audit; verify_references and reconcile_values for number/reference checking
• •Multi-agent: multiple CLI agents can share one session or spread across siloed RStudio windows

Key features

• ✓Tools: execute_r, execute_r_with_plot, execute_r_async (via callr), list/connect_session, read_file (parses .docx/.pdf), get_viewer_content, modify_code_section, insert_text
• ✓Reviewer Zero protocol: extract/verify/recompute/reference-check every statistical claim, plus preregistration audits and specification-curve robustness checks
• ✓verify_references (CrossRef, retraction flags), search_citations (OpenAlex), get_bibtex (doi.org)
• ✓checkpoint_session / restore_session / list_checkpoints, generate_notebook (Quarto), generate_codebook
• ✓R Best Practices Protocol, Multi-Agent Coordination Protocol, AI-Driven Data Annotation workflow

Caveats

• ⚠Supervised power tool: the agent executes arbitrary R in your live session; validate_code_security() is a regex blocklist, "a seatbelt, not a sandbox"
• ⚠Binding to 127.0.0.1 is not a security boundary (origin block on by default; session token opt-in and off by default for upgrade compatibility)
• ⚠Prompt-injection risk when auditing third-party manuscripts (untrusted content on a path to code execution)
• ⚠clauder-mcp pins mcp<2 (SDK 2.0 migration pending); one desktop app per session, though multiple CLI agents can share