WebMCP · 4 min read
What Is WebMCP?
How browser agents use website tools, with a supplier-catalog example, interface comparisons, and the limits to check before a pilot.
WebMCP lets a website expose tools that a compatible browser agent can call. A supplier might offer search_products or prepare_quote_request, with defined inputs and results. The agent can use those tools while the customer stays on the website.
It is a proposed standard, and browser support is still experimental. Chrome offers an origin trial; check the browser and agent your customers would actually use. The proposal and Chrome overview describe its current status.
A supplier-catalog example
A maintenance manager needs 20 filter cartridges. They know the filtration rating and dimensions, but not the supplier's part number. They ask an assistant: “Find the 10 micron cartridges, show me the dimensions, and prepare a quote for 20.”
On a site with suitable WebMCP tools, the agent could search the catalog, return matching products with specification links, and fill a quote request. The buyer checks the part and quantity before sending it.
The example is fictional. The tool would return published specifications; it would not establish that a cartridge fits a particular machine or that the supplier has stock. Those checks still need the appropriate information.
What the website adds
Without a tool interface, a browser agent has to interpret the page and work through its controls. It might choose the wrong field, miss a hidden selection, or mistake a prepared form for a sent request.
A WebMCP tool gives the action a name and defined inputs. The website runs its own application logic and returns the result. Developers can annotate an HTML form or register a JavaScript tool.
Both approaches can reuse the regular interface. They also depend on the underlying application: incorrect catalog data, broken validation, or missing permissions still need fixing.
Where it might help
A catalog search is one candidate. Others include finding an invoice in a customer portal, assembling a service request, or preparing an update in an internal application. These tasks often require users to carry information between several screens.
For a booking request, an assistant might gather dates, locations, and passenger details. The application would still check availability and eligibility. For an account change, it would enforce the customer's permissions and ask for any required confirmation.
Try one of these tasks with representative users and compare it with the existing interface. Check whether users reach the correct result with fewer errors or less effort.
How it relates to other interfaces
| Approach | What it provides |
|---|---|
| Website chatbot | A conversation that can answer questions or guide a visitor |
| API | Application operations for other software to call |
| Backend MCP | Tools for connected AI clients, beyond a single page visit |
| WebMCP | Tools available within a browser experience |
| Search optimization | Public information that search engines can discover and index |
A WebMCP tool can call an existing API, and a chatbot can use tools. Work that must continue without an open browser session may fit a backend connection better. Chrome's comparison of WebMCP and MCP explains that distinction.
Adding WebMCP does not automatically put a business in search results or ChatGPT answers. Google documents ordinary search requirements for its AI features. OpenAI documents search crawler access separately.
What a pilot needs to establish
Check which browsers and agents support your tools, and whether your customers use them.
Then test incomplete input, duplicate requests, rejected actions, and stale records. Make the visible page agree with the result returned to the agent. A filled form should say it is ready for review, not that it has been sent.
Enforce permissions in the application. Read-only tools can expose private data, and returned customer text may contain hostile instructions. Google's tool security guidance covers these concerns.
Finally, assign someone to maintain the tools as browsers and the application change. Keep the regular interface working for visitors without compatible agents.
Permadyn's public integration lets agents find services, use the project scoper, and prepare inquiries for on-page approval. It demonstrates that implementation pattern; we have not established a sales increase from it.
For a code example and testing steps, read how to make your website WebMCP ready. We also offer WebMCP implementation and an agent-ready software assessment.
Original technical status review: September 4, 2026. Chrome's overview was rechecked September 20, 2026.