01 · FoundationsAll levelsHalf day
WebMCP and the agent-ready web
Understand the shift from websites designed for people to browse towards websites that can expose structured actions to agents.
- What WebMCP is, what MCP is, and how they differ.
- How agents discover tools, understand inputs and invoke actions.
- What the browser, website and connected service each control.
- Current standards, browser experiments and implementation realities.
- Where the business opportunities and limitations sit.
You leave with: a shared vocabulary, a system map and a shortlist of real tasks worth making agent-ready.
02 · Product and businessLeaders and product teamsHalf day
Designing agent-ready customer journeys
Start with the customer or employee outcome, not the technology. Find the friction and design a better task.
- Map a journey from intent to verified outcome.
- Identify actions, decisions, hand-offs and failure points.
- Choose what belongs in the website, an MCP server or an existing API.
- Design human approval points and useful recovery paths.
- Prioritise opportunities by value, feasibility, risk and readiness.
You leave with: a prioritised agent-ready workflow blueprint and measurable acceptance criteria.
03 · Developer workshopDevelopers1 day
Build your first WebMCP tools
Move from the concept to a working browser-based implementation using a controlled example site.
- Prepare a WebMCP-capable development environment.
- Define tool names, descriptions, input schemas and outputs.
- Register and expose website actions using the current supported API.
- Discover and invoke tools from a compatible agent/browser setup.
- Validate inputs, handle errors and keep actions within site rules.
- Test the experience and document browser dependencies.
You leave with: a working read-only WebMCP example, source code and a repeatable test checklist.
04 · IntegrationDevelopers and architects1 day
WebMCP + MCP: connect the whole workflow
Understand how browser-facing actions connect to business systems without turning the website into an unrestricted integration layer.
- Separate browser tools, MCP servers, APIs and system-of-record responsibilities.
- Connect a website workflow to a controlled MCP service.
- Map identity, session context and least-privilege access.
- Handle asynchronous work, hand-offs and task continuity.
- Design idempotency, timeouts, retries and failure recovery.
You leave with: a reference architecture and a working end-to-end integration pattern using safe example services.
05 · Trust and securitySecurity, risk and productHalf day
Consent, identity and safe agent actions
Capability is not permission. Learn how to make agent activity understandable, bounded and accountable.
- Authentication versus authorisation, consent and delegated authority.
- Read-only discovery versus actions that change the world.
- Exact action scope, input validation and approval boundaries.
- Protect credentials, personal data and sensitive outputs.
- Record requests, approvals, execution receipts and verified outcomes.
- Stop safely when the result is uncertain; never blindly replay consequential actions.
You leave with: a threat and permission model, an approval matrix and an evidence checklist.
06 · Quality and operationsDelivery teamsHalf day
Test, evaluate and operate agent-ready services
Prove that the task works. A successful demo or green unit test is not the same as a verified real-world outcome.
- Test tool discovery, schemas, invocation and expected results.
- Evaluate task completion, reliability and accessibility.
- Test denial paths, expired sessions, timeouts and partial completion.
- Separate simulated, browser-tested and production evidence.
- Use receipts, telemetry and release checks to support operational confidence.
You leave with: an evaluation plan, acceptance suite and evidence-led release checklist.
WebMCP is an evolving browser capability. Practical exercises will clearly identify the browser/version and feature configuration used. Where WebMCP is unavailable, the underlying task remains usable through a conventional interface or documented fallback. No course promises universal browser support.