Use Console from R

Add Console to a chat

Install the package and its dependencies:

pak::pak("github::posit-dev/mcp-console/r")

Register a Console tool with your chat:

library(ellmer)
library(mcp.console)

chat <- chat_openai()
tool <- console_tool()
chat$register_tool(tool)
chat$chat("Explore mtcars with Console. Keep the data for follow-up.")

Your model provider’s credentials belong in the calling R process. Keep the tool available while the chat is using it; state persists while its Console connection remains alive. For dependent calls through chat$chat_async(), select tool_mode = "sequential".

console_tool() uses Console on PATH or launches the latest published release through reticulate::uv_run_tool(). The package guide covers selecting an executable or version.

Choose the workspace and settings

Pass project to select the directory where Console starts. Without an explicit config, Console discovers global and project configuration files. For example:

tool <- console_tool(project = "/path/to/project")

The directory must already exist. Relative paths in Console use this directory without changing the calling R session’s working directory.

To supply settings directly, pass a ConsoleConfig object:

config <- ConsoleConfig(sandbox = SandboxPolicy(profile = "workspace"))
tool <- console_tool(config = config, project = "/path/to/project")

An explicit configuration skips file discovery. Here the workspace profile allows ordinary project writes; network access is configured separately. Configure a Console session covers inspecting, editing, saving, and reusing settings. The R API documents the complete configuration interface.

Work with the session

The shared guides explain sessions and languages, packages and environments, sandbox permissions, and recordings and reports. These describe the same workbench your agent uses from R.

Build a data assistant shows how to change the descriptions, languages, and parameters the agent sees. For a complete analysis example, see Giving an agent an R session.