Sake.Tool behaviour (Sake v0.1.0)

Copy Markdown View Source

A tool: use Sake.Tool with options, arg lines, an optional output block, and execute/2.

defmodule MyApp.MCP.SearchIssues do
  use Sake.Tool, name: "search_issues", description: "Search issues", scope: "issues:read"
  arg :query, :string, required: true
  @impl true
  def execute(%{query: q}, ctx), do: {:ok, Issues.search(ctx.scope, q)}
end

Options: name (default: the module's last segment in snake case), title, description, icons, scope (one or many), annotations (read_only, destructive, idempotent, open_world), stream, ttl, meta, input_schema (a schema map instead of arg lines).

Summary

Callbacks

complete(arg, partial, ctx)
execute(args, ctx)
spec()

Functions

arg(name, type, opts \\ [])

Declares an argument (or, inside output, an output property).

output(list)

Declares the output schema with arg lines.

Callbacks

complete(arg, partial, ctx)

(optional) 
@callback complete(arg :: String.t(), partial :: String.t(), ctx :: Sake.Context.t()) ::
  {:ok, [String.t()]} | {:ok, [String.t()], keyword()}

execute(args, ctx)

@callback execute(args :: map(), ctx :: Sake.Context.t()) ::
  {:ok, term()} | {:ok, term(), keyword()} | {:error, String.t()}

spec()

@callback spec() :: Sake.Tool.Spec.t()

Functions

arg(name, type, opts \\ [])

(macro)

Declares an argument (or, inside output, an output property).

output(list)

(macro)

Declares the output schema with arg lines.