What is a Tool Registry in AI Agent Systems?

Think of a tool registry like an app store for AI agents. Instead of an agent knowing at build time exactly which tools exist, a registry provides a structured directory of available tools — each with a name, description, input schema, and invocation endpoint. When an agent needs to accomplish a subtask, it queries the registry to discover which tools are relevant, selects the best match, and invokes it. In simple systems, the tool registry is just a list of function definitions passed to the model in the system prompt. In production systems, it's a dynamic service: tools can be registered and deregistered at runtime, access permissions can be enforced per agent or user, tool descriptions can be versioned, and usage can be logged and monitored. The registry pattern solves a real scaling problem: as agentic platforms grow from 5 tools to 50 to 500, statically listing all tools in every prompt becomes unmanageable and wasteful. Instead, the agent retrieves only the relevant subset of tools for the current task — a capability called tool retrieval, which mirrors how RAG retrieves relevant documents. MCP (Model Context Protocol), Anthropic's open standard for tool integration, is an emerging standard for how tool registries expose capabilities to AI agents across organizational boundaries.