class RubyLLM::MCP::Tool

A tool offered by an MCP server. It works anywhere a Tool does: a chat renders its name, description, and schema for the model, and calling it calls the server.

tool = linear.tools.first
tool.read_only?  # => true
chat.with_tools(tool)

The behavior predicates read the server’s annotations. They are hints from the server, so trust them only as far as you trust the server.

Attributes

description [R]

The description the model sees.

name [R]

The name the model calls this tool by.

parameters_schema [R]

The JSON Schema for the arguments the model provides.

server_name [R]

The tool’s name on the server.

Public Instance Methods

call (**arguments)

Source

# File lib/ruby_llm/mcp/tool.rb, line 78
def call(**arguments)
  @mcp.run(self, arguments.except(:tool_call))
end

Calls the tool on the server and returns what the model sees: the result’s content, what the wrap: method made of it, or { error: } when the tool failed. Raises MCP::InputRequiredError when the server needs input that no MCP.before_input_request callback gave.

destructive? ()

Source

# File lib/ruby_llm/mcp/tool.rb, line 52
def destructive?
  !read_only? && @annotations['destructiveHint'] != false
end

Returns whether the tool may destroy or overwrite data. Tools that are not read-only count as destructive unless the server says otherwise.

idempotent? ()

Source

# File lib/ruby_llm/mcp/tool.rb, line 58
def idempotent?
  @annotations['idempotentHint'] == true
end

Returns whether calling the tool again with the same arguments has no further effect.

open_world? ()

Source

# File lib/ruby_llm/mcp/tool.rb, line 63
def open_world?
  @annotations['openWorldHint'] != false
end

Returns whether the tool reaches beyond the server, such as the web.

read_only? ()

Source

# File lib/ruby_llm/mcp/tool.rb, line 45
def read_only?
  @annotations['readOnlyHint'] == true
end

Returns whether the server says the tool only reads.

requires_approval? ()

Source

# File lib/ruby_llm/mcp/tool.rb, line 69
def requires_approval?
  @mcp.requires_approval?(self)
end

Returns whether the tool pauses for approval, as declared with MCP.requires_approval.