Add Context to Chat Messages

Formats R objects and combines them with optional text to provide context for chat messages. Similar to btw::btw(), this function wraps formatted output in a structured format that chat models can easily understand.

This function uses mcp_describe internally to format objects into human-readable text, then wraps the output in code blocks and returns an ellmer::ContentText object suitable for use with ellmer::Chat$chat().

Usage

mcp_context(..., .header = "## Context", .wrap_code = TRUE)

Arguments

...

R objects to format, or character strings to include. Each argument will be formatted using mcp_describe() unless it's a plain text string.

.header

Character string for the context header. Default: "## Context"

.wrap_code

Logical. Whether to wrap non-text objects in code blocks. Default: TRUE

Value

An ellmer::ContentText object containing the formatted context. This object can be passed directly to chat$chat().

Details

The function processes each argument as follows:

• Package references like "{packagename}" are expanded to show all help topics from that package in JSON format

• Plain text strings (single character vectors) are used as-is

• Other R objects are formatted using mcp_describe()

• Formatted objects are wrapped in code blocks if .wrap_code = TRUE

• All parts are combined with the header into a single text block

The resulting ContentText object is compatible with ellmer chat functions and will be automatically processed when passed to chat$chat().

See also

mcp_describe for the underlying formatting function, mcpflow_instantiate for workflow integration

Examples

if (FALSE) { # \dontrun{
# Single object
mcp_context(sessionInfo())

# Package help topics
mcp_context("{ravepipeline}")

# Multiple objects with question
mcp_context(sessionInfo(), mtcars, "What packages and data are available?")

# Use in chat
chat <- ellmer::chat_anthropic()
response <- chat$chat(
  mcp_context(sessionInfo(), "What R version am I using?")
)

# Combine with workflow instantiation
wf <- mcpflow_read("ravepipeline::rave_pipeline_class_guide")
chat <- mcpflow_instantiate(wf, chat = chat)
response <- chat$chat(
  mcp_context(
    sessionInfo(),
    "Generate code for power analysis"
  )
)
} # }