claudeR: R Interface to Anthropic's Claude

First Commit: April 29, 2023

This package provides two ways to interact with Claude from R:

1. API Interface (claudeR) - Direct API calls for text completions with streaming and extended thinking support
2. CLI Interface (claude_code) - Integration with Claude Code, Anthropic's agentic coding CLI for file operations, code execution, and multi-turn interactions

Supports Claude 2, Claude 3, Claude 3.7, and Claude 4.5 models (Sonnet, Opus, Haiku). Defaults to Claude 4.5 Sonnet.

Installation

# Install from GitHub
devtools::install_github("yrvelez/claudeR")

# For CLI features, also install Claude Code
# npm install -g @anthropic-ai/claude-code

Quick Start

library(claudeR)
Sys.setenv(ANTHROPIC_API_KEY = "your_api_key_here")

# API: Simple completion
claudeR(
  prompt = list(list(role = "user", content = "What is R?")),
  max_tokens = 100
)

# CLI: Agentic coding (requires Claude Code CLI)
claude_code("List the files in the current directory")

API Interface

Basic Usage

# Claude 4.5 Sonnet (default)
response <- claudeR(
  prompt = list(list(role = "user", content = "Explain tibbles in R")),
  max_tokens = 200
)

# Use a different model
response <- claudeR(
  prompt = list(list(role = "user", content = "Explain tibbles in R")),
  model = "claude-opus-4-5-20251101",
  max_tokens = 200
)

# With system prompt
response <- claudeR(
  prompt = list(list(role = "user", content = "Review this code: x <- c(1,2,NA); mean(x)")),
  system_prompt = "You are an expert R programmer. Be concise.",
  max_tokens = 300
)

Extended Thinking

For complex reasoning tasks, enable extended thinking:

result <- claudeR(
  prompt = list(list(role = "user", content = "Solve: If f(x) = 2x^2 + 3x - 5, find x where f(x) = 20")),
  thinking = list(type = "enabled", budget_tokens = 2000),
  return_thinking = TRUE
)

cat("Thinking:\n", result$thinking, "\n\n")
cat("Answer:\n", result$response)

Model Options

Model	ID	Use Case
Claude 4.5 Opus	claude-opus-4-5-20251101	Most capable, complex reasoning
Claude 4.5 Sonnet	claude-sonnet-4-5-20250929	Balanced (default)
Claude 4.5 Haiku	claude-haiku-4-5-20251001	Fast, cost-efficient
Claude 3.7 Sonnet	claude-3-7-sonnet-20250219	Extended thinking
Claude 2	claude-2	Legacy support

API Parameters

Parameter	Description	Default
prompt	String (Claude 2) or list of messages (Claude 3+)	required
model	Model identifier	claude-sonnet-4-5-20250929
max_tokens	Maximum tokens to generate	100
temperature	Randomness (0-1). Must be 1 with thinking.	0.7
system_prompt	System instructions (Claude 3+)	NULL
thinking	Extended thinking config	NULL
return_thinking	Return thinking in output	FALSE
stream_thinking	Stream response in real-time	TRUE

CLI Interface (Claude Code)

The CLI interface provides agentic capabilities—Claude can read/write files, execute code, and perform multi-step tasks.

Setup

# Check if Claude Code is installed
claude_code_available(verbose = TRUE)

# View configuration
claude_code_config_print()

# Configure settings
claude_code_config(timeout = 600)

The Claude Pipe Operator

mtcars %|c>% 
 "Generate a figure with mpg on x axis and wt on y axis" %|c>%
 "Facet by cylinders" %|c>%
 "Maximize data-ink ratio"

Basic Usage

# Simple prompt
claude_code("What files are in the current directory?")

# With JSON output
result <- claude_code("List 3 R packages for visualization", output_format = "json")

# Restrict available tools
claude_code("Analyze data.csv", allowed_tools = c("Read", "Bash"))

Convenience Functions

# Pipe data to Claude
mtcars  %|c>% "Describe this dataset and suggest visualizations"

# Analyze a file
claude_code_file("analysis.R", "Review this code for potential issues")

# Code review with focus
claude_code_review("script.R", focus = "performance")

# Generate code
claude_code_generate("function to calculate rolling mean", language = "R")

# Analyze a data frame
claude_code_analyze(iris, "What patterns exist in this data?")

Batch Processing

prompts <- c(
  "Explain ggplot2",
  "Explain dplyr",
  "Explain tidyr"
)

results <- claude_code_batch(prompts, progress = TRUE)

Session Management

# List previous sessions
claude_code_sessions()

# Resume a session
claude_code_session("Continue our analysis", session_id = "last")

Interactive Chat

# Start an interactive chat (in R console)
claude_code_chat()

Response Utilities

response <- claude_code("Write an R function to parse JSON")

# Extract code blocks
code <- claude_code_extract(response, language = "r")

# Parse JSON from response
data <- claude_code_parse(response)

# Pretty print
claude_code_print(response)

CLI Parameters

Parameter	Description	Default
prompt	The prompt to send	required
output_format	"text", "json", or "stream-json"	"text"
model	Model to use	config default
max_turns	Maximum agentic turns	NULL
system_prompt	Custom system prompt	NULL
allowed_tools	Tools Claude can use	all
disallowed_tools	Tools Claude cannot use	none
permission_mode	"default", "acceptEdits", "bypassPermissions"	"default"
working_dir	Working directory	getwd()
timeout	Timeout in seconds	300

When to Use API vs CLI

Use Case	Recommended
Simple text completion	API (claudeR)
Streaming responses	API (claudeR)
Extended thinking	API (claudeR)
File operations	CLI (claude_code)
Code execution	CLI (claude_code)
Multi-step tasks	CLI (claude_code)
Interactive coding	CLI (claude_code_chat)

License

MIT License