Foundations: how agents use a website today|INEMA.CLUB-PRO

WebMCP Program · course 0 of 6 · 8 lessons

Foundations: how agents use a website today, and what your site shows them

Reading, operating the screen, and calling a tool: the three ways an agent can use your site. You see the accessibility tree it receives, fix your HTML, place MCP and WebMCP, read the draft's status with its date, and leave with a diagnosis of your site. All local, no quota spent.

Start with lesson 1 See the WebMCP Area
Dark monitor with a browser open on a form and a cable plugged into a connector beside it.
  • 8lessons
  • 21exercises with expected output
  • 7scripts in the kit, all local
  • 0model calls

The lessons

From what the agent does today to your site's plan: three ways, the tree, the HTML, MCP and WebMCP, stable vs. draft, who shows up, the diagnosis, and the plan.

  1. Lesson 1 · 15 minThree ways an agent can use your site: read, operate the screen, call a toolWhen an AI agent reaches your site, it does one of three things: it reads the HTML, it operates the screen like a person, or it calls a tool the site has declared. In this lesson you run all three against a sample page and against a real site, and see which one works in each case.
  2. Lesson 2 · 15 minWhat the agent sees: accessibility tree × screenshotYou see colors, a yellow button, and a date field. A browser agent like Playwright MCP gets something else: a list of roles and names, the accessibility tree. In this lesson you print that tree for the example page, compare it with the screenshot, and learn to read what's missing.
  3. Lesson 3 · 20 minSemantic HTML and a well-labeled form: the interface every agent already readsBefore any tool for agents, there is an interface they all already read: HTML. A label tied to its field, a real button, a main region, a heading, a language, and alt text. It is stable, it works for screen readers and for agents, and it does not depend on an experimental browser. In this lesson you fix the sample page until the check comes back at zero.
  4. Lesson 4 · 15 minMCP and WebMCP in one picture: server × page, who calls whomThe two names look like the same thing, but they are not. MCP connects an agent to a server of yours, outside the browser, and it is stable. WebMCP lets the page itself declare tools for the agent running in the browser, and it is a draft. In this lesson you separate the two in one picture and run the feature detection that keeps the site working without the API.
  5. Lesson 5 · 15 minStable × draft: reading the spec, Chrome Status, and the browsers' positions, with datesWebMCP has already changed its name once and may change again before the origin trial ends. Whoever teaches or implements it needs to know, with dates, what the spec says, which browser runs it, and who is against it. In this lesson you read all of that straight from the primary sources with a script, and you learn to separate what is foundation from what is lab work.
  6. Lesson 6 · 15 minWho's arriving: bots by purpose and signed agents (the concept)Not every AI bot that visits your site does the same thing: some index for search, others read when a person asks, others collect for training. The robots.txt file asks by purpose, but it doesn't prove who is who. In this lesson you read your robots.txt by purpose and learn what the Web Bot Auth draft proposes for identifying the agent.
  7. Lesson 7 · 20 minThe WebMCP Readiness diagnosis: run it and read it without falling for "high score = ranking"WebMCP Readiness is INEMA's open scanner: it opens the page in an isolated browser, only reads, and returns a score in four categories. In this lesson you run the scanner on your machine, with a pinned version, against the example page and against your own site, and you learn to read the result as a diagnosis, not as a promise of ranking.
  8. Lesson 8 · 20 minYour site's plan for agentsYou've already run the three ways, read the tree, fixed the HTML, placed MCP and WebMCP, read the dated state, checked who gets in, and run the scanner. Now you put it all into a one-page plan, with dates, and check that it's complete. This is the plan you take to the next courses in the program.

Who it's for

  1. 1Web devs, makers, or technical site owners who edit HTML and JavaScript and run commands in the terminal.
  2. 2People who want AI agents to use their site, in the browser and through MCP, and have never worked on this.
  3. 3People who have heard of WebMCP and want to know what is stable and what is still a draft.

What you'll be able to do

  1. 1Run the three ways an agent can use a site against your page and see which one works.
  2. 2Read the accessibility tree the agent receives and bring the HTML problems down to zero.
  3. 3Say, with a date and a source, where WebMCP and the browsers' positions stand.
  4. 4Run WebMCP Readiness on your machine and build your site's plan for agents.

Where this course fits in the WebMCP Program

  1. 1This is course 0: the foundation for the other five. Suggested order: 0 → 1 → 2 → 3 → 4 → 5.
  2. 2After it: 1 · MCP server for your service, 2 · WebMCP on the page, 3 · browser agents, 4 · security, 5 · expert.
  3. 3Before you start: Node 20 or newer, git, and Python 3. Linux or macOS; on Windows, use WSL.
  4. 4The kit, the sample page, and the script messages are in Portuguese in all versions of the course; the commands are the same.