Claudio.MCP.ResultMapper (Claudio v0.5.0)

Maps tool use blocks from Claudio responses back to MCP call format.

After Claude responds with tool_use or mcp_tool_use blocks, this module extracts them into a format suitable for calling back to MCP servers.

Example

response = Claudio.Messages.create(client, request)
calls = Claudio.MCP.ResultMapper.extract_mcp_calls(response)

results = Enum.map(calls, fn call ->
  {:ok, result} = MyAdapter.call_tool(mcp_client, call.name, call.arguments)
  Claudio.Tools.create_tool_result(call.id, result)
end)

Summary

Functions

claudio_to_mcp(response)

Alias for extract_mcp_calls/1.

extract_mcp_calls(response)

Extracts MCP tool calls from a response.

extract_mcp_calls(response, server_name)

Extracts MCP tool calls for a specific server.

Types

mcp_call()

@type mcp_call() :: %{
  id: String.t(),
  name: String.t(),
  arguments: map(),
  server_name: String.t() | nil
}

Functions

claudio_to_mcp(response)

@spec claudio_to_mcp(Claudio.Messages.Response.t()) :: [mcp_call()]

Alias for extract_mcp_calls/1.

extract_mcp_calls(response)

@spec extract_mcp_calls(Claudio.Messages.Response.t()) :: [mcp_call()]

Extracts MCP tool calls from a response.

Handles both mcp_tool_use blocks (from server-side MCP connector) and regular tool_use blocks with prefixed names (from client-side MCP tools).

Note: Regular tool_use blocks are identified as MCP calls when their name contains __ (double underscore) as a separator — e.g., server_name__tool_name. If you define non-MCP tools with __ in their names, they will be incorrectly captured. Use mcp_tool_use blocks (from the server-side MCP connector) or avoid __ in regular tool names to prevent ambiguity.

extract_mcp_calls(response, server_name)

@spec extract_mcp_calls(Claudio.Messages.Response.t(), String.t()) :: [mcp_call()]

Extracts MCP tool calls for a specific server.