OpenAI integrated Web Model Context Protocol (WebMCP) support into the built-in browser of its ChatGPT desktop application, allowing websites to expose specific actions directly to AI agents.
This new functionality enables developers to define tools within their webpages that ChatGPT can utilize, streamlining interactions such as searching documents, editing files, or updating shopping carts without requiring separate connectors.
WebMCP operates directly within an open webpage, linking to the user’s current signed-in session, which distinguishes it from the server-based Model Context Protocol (MCP), according to OpenAI.
Developers can register JavaScript functions as tools, providing names, descriptions, and structured input schemas that define how AI agents can interact with their applications.
The site tools are available in the ChatGPT desktop app for accounts using GPT-5.6 Sol or Terra models, OpenAI stated, adding that the feature is not active for GPT-5.6 Luna users and is disabled in Enterprise or Edu workspaces.
Before ChatGPT interacts with a site using these tools, OpenAI requires explicit user permission. The company also confirms actions that would modify data or share personal information, adding a layer of security.
The WebMCP specification is currently a draft from the W3C Web Machine Learning Community Group and has not yet advanced to the W3C Standards Track, indicating it is still under development.
OpenAI conducts security checks for each tool invocation. However, the company cautioned about potential risks such as data exfiltration and prompt injection, advising developers to implement strong security practices.
Source: Search Engine Journal
