Skip to content

Template Classes & Registration

Why: Maximum control over tool definitions, metadata, validation, and execution lifecycle. Jump to symbols: API Exports Index

Exports: BaseTemplate, FunctionToolTemplate, MCPToolTemplate, CodeExecToolTemplate, AgentTemplate, register_template

BaseTemplate

  • What: Abstract base for all tool templates.
  • When: Subclass to build custom behaviors beyond built-ins.
  • How: Define name, description, parameters, and run().

FunctionToolTemplate

  • What: Define function-style tools with explicit schema.
  • When: Prefer when you need strong schema control or custom validation.
  • How: Set parameters with ToolParameter entries; implement run().

Example:

from orchestrator import FunctionToolTemplate
from orchestrator.shared.models import ToolParameter

class AddTool(FunctionToolTemplate):
    name = "add"
    description = "Sum two numbers"
    parameters = [
        ToolParameter(name="a", type="number", required=True),
        ToolParameter(name="b", type="number", required=True),
    ]
    async def run(self, a: float, b: float) -> dict:
        return {"sum": a + b}

MCPToolTemplate

  • What: MCP-flavored template with richer discovery metadata.
  • When: Use when exposing tools to MCP-based planners/adapters.

CodeExecToolTemplate

  • What: Tools that execute user-provided code with sandbox controls.
  • When: Use for safe code execution tasks; pairs with sandbox features.

AgentTemplate

  • What: Define agent-style tools (evaluators, classifiers, planners).
  • When: Provide consistent schema and lifecycle around agent calls.

register_template()

  • What: Add a custom template to the registry.
  • When: Make your custom template discoverable for use.
  • How: Call with your class reference.
from orchestrator import register_template
register_template(AddTool)

Related: - Sample: samples/23-adding-new-tools/three_ways.py