Tool Registry

Hardcoding tools in every agent file doesn't scale. A registry enables dynamic registration, discovery, versioning, and permission control.

A tool registry is a hardware store inventory system — every tool has a SKU, description, and location, and the clerk (agent) looks up what to fetch.

Scenario

A registry holds 20 tools but the 'read-only analyst' agent only receives search_docs, run_sql_query (SELECT only), and generate_chart.

Solution

In Tool Calling & Function Calling, apply Tool Registry to this scenario: A registry holds 20 tools but the 'read-only analyst' agent only receives search_docs, run_sql_query (SELECT only), and generate_chart. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Tool Registry (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Tool Registry happens in the code.

1class ToolRegistry:  # define a data structure or component2    def __init__(self):  # define a reusable function3        self._tools = {}  # key line for Tool Registry4    def register(self, name, schema, handler, roles=None):  # define a reusable function5        self._tools[name] = {"schema": schema, "handler": handler, "roles": roles or ["*"]}  # key line for Tool Registry6    def get_schemas(self, role="*"):  # define a reusable function7        return [t["schema"] for t in self._tools.values() if role in t["roles"]]  # return the result8    async def execute(self, name, args):9        return await self._tools[name]["handler"](**args)  # return the result

• Treating Tool Registry as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for tool registry

Quick recap — the most important points from this module.