About the Seminar
This seminar covers building working agentic automations by combining three layers: Claude as the reasoning engine, MCP (Model Context Protocol) as the standard way to give Claude access to tools and data, and n8n as the workflow orchestrator that triggers, schedules, and connects everything to real business systems.
The day is hands-on. Participants build a complete automation end to end: a trigger fires in n8n, Claude reasons over the input, calls MCP tools to fetch and act on real data and returns a result that is written back to a business system.
Who Is This Seminar For?
- Automation engineers and integration developers
- IT and operations staff who own internal workflows
- Solution architects evaluating agentic patterns
- Technical product owners who will scope agent projects
Prerequisites
- REST APIs: endpoints, headers, JSON request/response
- Basic scripting in any language (JavaScript or Python preferred)
- Understanding of authentication basics: API keys, tokens, environment variables
- General familiarity with LLM chat interfaces (prompting, context)
Key Topics
Framing Agentic Automation
- Deterministic workflows vs. agentic workflows
- Where agents add value and where they add risk
- The three-layer architecture: reasoning, tools, orchestration
- Walkthrough of the automation built during the day
Claude as the Reasoning Layer
- The Messages API: system prompts, context, and structured output
- Tool use fundamentals and the tool-call loop
- Prompting for reliability: constraints, schemas, refusal paths
- Lab: a Claude call that returns validated structured JSON
MCP - Giving Claude Real Access
- What MCP is and the problem it standardizes
- Servers, clients, and transports (stdio vs. remote)
- Tools, resources, and prompts as MCP primitives
- Installing and configuring an existing MCP server
- Lab: connect a filesystem and an HTTP MCP server to Claude and drive them by prompt
- Permissions, scopes, and what not to expose
n8n as the Orchestration Layer
- n8n concepts: nodes, triggers, credentials, executions
- Triggers: webhook, schedule, and app events
- Data shape in n8n: items, expressions, and mapping
- Calling Claude from n8n: AI nodes vs. plain HTTP request
- Lab: a webhook-triggered workflow that classifies an incoming request with Claude
End-to-End Build
- Designing the pipeline: trigger → reason → act → write back
- Wiring MCP-backed tool access into the n8n execution path
- Branching on Claude's decision and handling the unhappy path
- Human-in-the-loop approval steps
- Lab: complete the full automation and run it against live data
Expected Outcomes
- A working agentic automation running locally on the participant's machine
- A reusable pattern for connecting Claude to internal tools via MCP
- A checklist for evaluating whether a given process should be automated agentically
- Reference workflow templates and configuration files to take back to work