Meeting users where they are: why we built MCP

This doc explains why Ethics of AI exposes a Model Context Protocol (MCP) surface alongside the website. It is about the user-engagement problem MCP solves, not the implementation. The companion piece, How we built a remote MCP server for Ethics of AI, covers the technical choices.

The browser-only assumption

A traditional SaaS product treats the browser as the user's home. You ship a website, put the work behind a sign-in, and wait for users to come to your URL.

One user, one browser, one path in. The browser fetches pages, holds the Clerk session, and turns clicks into API calls. The implicit message is "come to us": every interaction routes through that single column. That assumption used to be solid. It is starting to leak.

Users are not single-surface

A real user rarely sits in one tool all day. The same person who buys a certification on the website may write assessment notes in Claude Desktop an hour later, draft a response in claude.ai during a meeting, or ask Claude on their phone what their next step is on the way home. The AI client is not a side surface for these users; it is where the working day actually happens.

If the platform only speaks browser, every non-browser surface is a dead end:

To engage from a non-browser surface, the user has to leave the conversation, open a browser, sign in, find the page they want, do the thing, and come back. Every one of those steps is a chance to drop the task. For an assessment-shaped product, that drop is the whole funnel.

The fix is not a better browser experience. The fix is to stop forcing the context switch.

MCP as a second front door

MCP is a protocol that AI clients understand. It lets a server expose tools and resources the client can call on the user's behalf, with the user's consent and identity. By running a remote MCP server, Ethics of AI gains a second front door, sitting alongside the website:

The two surfaces mirror each other. The website is a user agent that authenticates with Clerk and calls the Users API; the AI client, via MCP, is also a user agent that authenticates with Clerk and calls the Users API. Same human, same identity, same data, same authorisation policy. The only thing that changes is which surface the user starts from.

Neither surface is promoted over the other. The website is where most users still discover Ethics of AI; the MCP surface is where the work increasingly happens for users who already live inside an AI client.

What this looks like in practice

A "second front door" lets a user do platform things without leaving the conversation they were already having:

The user asks for something in plain language. Claude picks the right MCP tool, sends the request, and shows the result back in the same conversation. No tab opened, no second sign-in, no context switch. The platform shows up where the user is and steps back out when the task is done.

Try it yourself

Add the Ethics of AI connector to your Claude client and start asking questions about your account in plain language. Setup is five clicks on Claude Desktop, with the same connector working on claude.ai and any mobile Claude app signed in to the same Anthropic account. You will need an Ethics of AI account with a purchased certification first, since the connector only acts on work tied to your account. See Connect your AI client for the step-by-step guide.

What this is not

Adding MCP is not a bet against the website. The browser is still the right surface for marketing pages, long-form articles, the dashboard, and any interaction that benefits from rich layout.

It is also not a parallel identity or authorisation model. Both surfaces use the same Clerk session and call the same Users API. MCP adds a transport, not a second policy.

What MCP is, is an acknowledgement that "user agent" is a bigger category than it used to be. A browser is a user agent. An AI client is a user agent. The platform itself is just the data and the policy underneath; the user agent is whatever the human happens to be using right now.

Takeaway

Humans use AI clients now. AI clients speak MCP. The data and the policy underneath should be reachable through whichever surface the human chose first.