Diagnosing a site with a critical eye
Read the scanner's WebMCP, SEO, GEO, and AEO scores and separate observable evidence from what depends on the backend and human review.
The problem
The next visitor to your site may not be a person.
Agents already open pages, fill in forms, and complete tasks. Without preparation, the site becomes a problem for them and for you.
- The agent operates the screen made for people and breaks when the layout changes.
- Nobody knows whether the reservation was made: the agent assumed it worked.
- The site has no robots.txt, sitemap.xml, or llms.txt, and the agent doesn't even decide to visit.
- A tool exposed without human confirmation changes data without anyone approving it.
- The security rule lives inside the prompt, where the model can ignore it.
After the training
By the end of the training, you can:
Turning forms into tools
Declarative API with toolname and description, and imperative JavaScript tools with JSON Schema, cancellation, and fallback.
Designing a coherent catalog
Good tools, state, and errors with recovery, migrating existing sites, and integration with modern frameworks.
Building the agent that uses the tools
Discovery, validated execution, cross-origin and permissions, and an observable conversational loop with authorization kept outside the model.
Taking it to production
Threat modeling, hybrid WebMCP + backend + MCP architecture, evals, observability, rollout, and governance.
5 courses in order · 16 modules · 96 topics
An overview and four phases, organized by the work you need to do.
Each phase has four modules of six topics. Site owners who don't code do the overview and the diagnostic and hand the phases to their team. Those who build continue from Builder to Expert.
- 0
Overview
WebMCP Training — Sites and Agents from Zero to Expert
The prerequisites, how the four phases connect, the final project, and the readiness diagnostic for your site.
- 1
Building
WebMCP Builder
WebMCP, MCP and the agentic web, development environment, experimental declarative API and imperative API, with a catalog validator.
- 2
Integrate
WebMCP Integrator
Tool design, state, errors and recovery, migrating existing sites, and modern frameworks without duplicating rules.
- 3
Orchestrate
WebMCP Agent Developer
Browser agents that discover and run tools, with cross-origin, permissions, and an observable conversational loop.
- 4
Operate
WebMCP Expert
Agent and tool security, WebMCP + backend + MCP, evals and observability, production and governance.
Who it's for
Made for you if
- You build sites or apps and want AI agents to use the site's functions without guessing the screen.
- You lead a development team and need to decide what to expose, with what confirmation, and with what permission.
- You own a site and want to know today, with evidence, what is missing for agents to arrive.
- You already use MCP with agents and want to understand where WebMCP fits in the browser.
Not for you yet if
- You first want to be found on Google, Bing, and AIs: start with the inemaSEO area.
- You are looking for a closed, stable standard: WebMCP is still experimental, behind an Origin Trial.
Included in the training
You measure your site before and after.
WebMCP Readiness, INEMA's free scanner, gives an overall score, four independent scores, and up to 12 prioritized fixes, with the courses recommended for each gap. It comes together with the projects and videos on the topic.
Projects and kits
Videos
Why this training
What changes compared to reading the specification on your own.
Diagnostic before study
You run the free scan, note the baseline, and join the phase that matches the gap.
Eight layers, each with an owner
Discovery and tools in Builder, contract and state in Integrator, control and permissions in Agent Developer, backend and evals in Expert.
Security outside the model
Human confirmation before mutations, real authorization on the backend. The LLM chooses the tool; the site decides whether it runs.
Honesty about the experimental
The program shows the limitations of the environment and separates what the scan proves from what requires human review.
Questions
Before you start.
Do I need to follow the order?
The order is overview → Builder → Integrator → Agent Developer → Expert. Site owners who don't code do the overview and the diagnostic and hand the phases to the team.
Do I need to know how to code?
For the four phases, yes: they use JavaScript and JSON Schema. The overview and the diagnostic require no code.
Does WebMCP already work in browsers?
It is an experimental technology, currently behind an Origin Trial. The program keeps the limitations visible so you can learn early, at low cost.
How big is the training?
Five courses. Each of the four phases has 4 modules of 6 topics: 16 modules and 96 topics in total.
Does WebMCP replace MCP?
No. WebMCP lives on the page; MCP handles what stays outside the browser. The Expert phase teaches the hybrid architecture without duplicating rules.
The best first module is your own site.
Run the diagnostic, review the evidence, and start with the overview already knowing which decisions you need to master.