Is the OpenAI Plugins repository still worth looking at? Understanding the evolution of tool calls from the plug-in era

Sat, 06 Jun 2026 22:26:00 +0800

openai/plugins is a sample repository related to early OpenAI Plugins. Although today’s OpenAI tool system has evolved into function calling, tools, Agents SDK, MCP, etc., this warehouse still has reference value: it can help understand how the matter of “letting the model call external tools” was first commercialized.

To today’s developers, it looks more like historical data and design samples than a go-to template for new projects.

Why is it still worth watching?

The core issues of AI tool invocation have not changed:

How does the model know which tools are available;
How to describe tool capabilities;
How to declare parameters;
How the results are returned;
How to handle permissions and user confirmation;
How third-party services can be safely called by models.

In the Plugins era, manifest, OpenAPI schema, server-side API and other methods are used to solve these problems. Later function calling, tool calling, and MCP are still evolving around these issues in essence.

Suitable for whom to watch

This warehouse is suitable for:

People who want to understand the historical design of OpenAI Plugins;
People working on AI tool ecosystem or Agent platform;
People who need to understand manifest / OpenAPI tool description;
People who want to compare the differences between Plugins, function calling, and MCP;
People who maintain old plugins or migrate old solutions.

If you are a new project, it is not recommended to directly copy the old Plugins model. Priority should be given to the current official API, tools, Apps SDK or MCP related documents.

What can be learned from this?

The most valuable thing to learn is not a certain piece of code, but the tool interface design idea:

The tool description must be understandable by the model;
API schema should be clear enough;
The returned results must be suitable for the model to continue reasoning;
User authorization cannot be hidden in a black box;
Have understandable error messages when external tools fail.

These principles are still valid today.

Summary

openai/plugins is more like slicing in the early stages of an AI tool call. It is not necessarily suitable for direct use in new projects, but it is very suitable for understanding how the line of “plug-ins, tools, and agent capability expansion” develops.

If you are currently integrating AI Agent tools, reading it can help you see the source of many design problems.

Reference sources