10 WebMCP resources worth reading

These ten resources cover the proposal, Chrome's implementation, browser runtimes, server-side MCP, and the current ChatGPT surface documented by OpenAI. Read them in this order.

1. W3C WebMCP draft

Why read: This is the source of truth for the proposed browser API: document.modelContext, tool registration, discovery, execution, permissions, and security.

Who it helps: Founders and engineers deciding what WebMCP is allowed to mean, browser-platform engineers, and anyone writing a compatibility layer.

Authority and current status: Primary W3C Web Machine Learning Community Group material. The live page is a Draft Community Group Report dated September 17, 2026, and explicitly says it is not a W3C Standard or Standards Track document.

2. When to use WebMCP and MCP

Why read: It answers the recurring question: WebMCP does not replace MCP. It compares lifecycle, connectivity, UI ownership, discovery, and use cases, then explains how a durable backend and contextual browser layer can coexist.

Who it helps: Founders, product leads, and architects choosing a page or service integration.

Authority and current status: Primary Chrome for Developers guidance. Published March 11, 2026 and last updated May 19, 2026; still presented as guidance for a proposed browser API.

3. Imperative API

Why read: This is the practical JavaScript path for dynamic tools. It covers registerTool, schemas, annotations, cancellation, discovery, execution, and cross-origin iframe controls.

Who it helps: Frontend engineers building stateful web apps.

Authority and current status: Primary Chrome implementation documentation. The page is marked Explainer, Origin trial, and Intent to Experiment; published May 18, 2026 and last updated September 11, 2026.

4. Declarative API

Why read: It shows how existing HTML forms can become structured tools through attributes such as toolname and tooldescription. That makes it a useful progressive-enhancement path for teams whose workflows already live in semantic forms.

Who it helps: Teams with existing support, booking, search, or account forms.

Authority and current status: Primary Chrome documentation. Published May 18, 2026; the page is linked to an Origin trial and Intent to Experiment, and says WebMCP remains under active discussion.

5. WebMCP best practices

Why read: This is the most useful bridge from demo to product work. It covers tool strategy, avoiding overlapping tools, clear names and descriptions, schema design, strict code validation, graceful failure, evals, and registration lifecycle.

Who it helps: Teams deciding which tools to expose and how to measure reliability.

Authority and current status: Primary Chrome guidance. Published and last updated May 18, 2026; still framed as Origin trial / Intent to Experiment material.

6. Build your user's agentic workflows with WebMCP tools

Why read: It gives a product-shaped method: define the goal and initial state, role-play the interaction, handle variance, fail with recovery guidance, then evaluate and use telemetry.

Who it helps: Founders and product designers choosing which tools to expose.

Authority and current status: Primary Chrome explainer. Published August 26, 2026, with links to the Origin trial and Intent to Experiment; the page says the API can change.

7. Join the WebMCP origin trial

Why read: This is the clearest browser-availability snapshot in Chrome's own words. It explains that the Chrome 149 origin trial is time-limited early access for live testing and feedback, rather than a stable baseline feature.

Who it helps: Teams testing Chrome's implementation and separating experiments from production promises.

Authority and current status: Primary Chrome blog post. Published June 9, 2026; it describes WebMCP as an origin trial and early-preview program feature.

8. Using site tools in the ChatGPT desktop app

Why read: It is the reality check for ChatGPT. OpenAI documents Site Tools in the desktop app's built-in browser, says access depends on account and model support, and says they are not currently available in Chrome. It also explains safety prompts.

Who it helps: Founders planning ChatGPT reach and QA teams testing a WebMCP site.

Authority and current status: Primary OpenAI Help Center documentation retrieved September 19, 2026. Treat it as current product documentation, not as a promise that every account, model, browser, or agent surface has the same support.

9. MCP-B WebMCP explanation and docs

Why read: MCP-B's docs explain where its compatibility layer ends and the WebMCP proposal begins.

Who it helps: Teams needing browser compatibility, React helpers, local testing, or desktop relays.

Authority and current status: Primary MCP-B project documentation; retrieved September 19, 2026.

10. MCP TypeScript SDK v2

Why read: WebMCP does not replace the server layer. This SDK shows how to build an MCP server with tools, resources, prompts, and stdio or web transports.

Who it helps: Backend engineers building a durable service across AI hosts and jobs.

Authority and current status: Primary MCP project SDK documentation. The current v2 page calls itself the stable release line implementing the July 28, 2026 MCP specification; retrieved September 19, 2026.

The pattern is straightforward: use the W3C draft for the browser surface, Chrome's docs for the current experiment, the compatibility docs for browser runtimes, and the MCP SDK for the durable service layer. Recheck browser and ChatGPT status before choosing a deployment target.