WebMCP Gives Your SaaS a Second User

Your SaaS is about to 2x its users.

Not by doubling its customer base. Every person who opens your product may soon bring a second user with them: their agent.

Most SaaS products are still built for one kind of user. A person sees a screen, understands the interface, and knows what to click.

An agent works differently. It may inspect the page, read the DOM, take screenshots, and simulate clicks. That can get a job done, but it is slow and fragile. Change a button label or move a field, and the agent has to figure the product out again.

WebMCP proposes a better way. A product can expose structured actions that an agent can discover and call in the live browser. Instead of guessing what a button does, the agent gets a defined tool with a name, description, and inputs.

For example, a customer success product could expose actions such as find_accounts_with_usage_drop, group_by_owner, and create_follow_up_tasks. The agent can run them through the product's own logic, while the user sees the results in the same interface.

For the parts of a product that expose these tools, agents no longer have to rely on DOM scraping and simulated clicks. They can use the application as the product team intended.

WebMCP brings agents into the product

Server-side MCP and WebMCP solve different problems.

Server-side MCP lets an agent connect directly to a service's backend. It is useful for work that does not depend on the current browser session or visible product state.

WebMCP works inside the live product. The agent has access to what the user is looking at, and the product can show what the agent did. That matters when a workflow needs context, review, or a human decision before the next step.

The WebMCP proposal is clear on this point. It is meant to complement backend MCP, not replace it. Its focus is a human and an agent working together in the browser. The explainer calls out shared context and user control as core goals.

Tools are not the agent

Exposing actions does not make a product agentic. It gives an agent a better way to act.

The harder question is what the agent should do. Which accounts deserve attention? What counts as a risky change? When should it prepare a recommendation, and when should it act?

A useful agent needs the product's context, rules, and memory. It needs to understand the work, not merely operate the interface. That is why a generic chat box is rarely enough.

LogicPair helps SaaS companies build that layer: a product-native agent that understands the customer's work, remembers what happened, and moves the right workflow forward.

Design for the job, not every button

The first move is not to expose every screen action as a tool. Pick a job that already takes too many steps and depends on product context.

It might be triaging a queue, investigating an exception, preparing a renewal follow-up, or configuring a report. Define the outcome first. Then expose the actions that let an agent help complete it.

Be precise about consequences. An action that sends an email, changes billing, deletes data, or shares customer information needs clear permissions and a clear confirmation path. The WebMCP specification calls out those risks because a browser agent may operate in an authenticated session. Read the security considerations.

That is not a reason to keep agents out of the product. It is a reason to make the product the place where the right rules and review happen.

It is early, but the product question is here now

WebMCP is still an experimental proposal. Chrome currently offers it through an origin trial and a testing flag, and the API will change. Chrome's implementation guide is explicit about that.

Still, it is worth paying attention. OpenAI is already running a WebMCP Challenge around apps that get better when people and agents use them together.

The web is gaining a new kind of user. The SaaS companies that benefit will not be the ones that expose the most tools. They will be the ones whose agents understand the job, work through the product's rules, and leave the customer in control.