Agents, Tools, and MCP, oh my! Next-level AI concepts for developers

AI is evolving fast, and so are the ways developers can integrate it into tech systems. A flurry of new approaches and tools surfaces every week, and it’s hard to know where to focus. In this session, we will pull back the curtain on the next wave of AI development with agents, tool integrations, and Model Context Protocol (MCP). We will break down what AI agents are, how they interact with tools and APIs, and why context is critical for building smarter, more reliable applications. Next, we will look at MCP and how it standardizes communication between AI models and external systems. Along the way, we will touch on related concepts and step through code and demos, giving you a complete roadmap to level up your AI skills. You won’t need a yellow brick road to follow along, but you will discover some magical new tricks to level up your AI skills!
Code: https://github.com/JMHReif/agents-mcp-ohmy

Transcript

1. Jennifer Reif jennifer.reif@neo4j.com @JMHReif github.com/JMHReif jmhreif.com linkedin.com/in/jmhreif/ Agents, Tools, and MCP, oh my! Next-level AI concepts for developers
2. About Jennifer Developer advocate, Neo4j • Java champion • Builder, coder, tech learner • Tech speaker/blogger/podcaster/author • Other: reader, mom, cat mom, etc Jennifer Reif jennifer.reif@neo4j.com @JMHReif github.com/JMHReif jmhreif.com linkedin.com/in/jmhreif/
3. The AI Stack in 2026 More != Better LLMs Vector search / naive RAG / easy RAG Advanced RAG / GraphRAG Agentic work fl ows / Agents Evals / Guardrails / Security Memory / State / History
4. Layers of AI More != Better • Complexity vs value
5. Act 1: Overture
6. The “Magic” of AI? • Fix-all silver bullet • Add AI = problems solved • Implementation isn’t so magical
7. Reality of Building with AI Can collapse under their own weight • Works in demo • Hallucinated API calls • Fragile integrations • Complexity overwhelm • Optimization nightmares
8. Demo: LLM alone Lacks speci fi c context and personalization
9. Act 2: Agents + Tools
10. What is an AI Agent? Reasoning loop 1.Receive input 2.Reason 3.Select tool 4.Execute 5.Observe result 6.Respond
11. Separate the responsibilities Traditional tech + GenAI • LLM -> decision • Agent -> coordination • Application -> executes
12. AI Agents… Plan, reason, and execute tasks for users • Four Key Components: • Model: reason, plan, respond • Tools: fetch, act, call • Orchestration: remember, track • Runtime: execute
13. Phase 1: Prompt Tricks Hope is not a strategy! • “Call this API by returning JSON…” • Problems: • Brittle • Hallucinated parameters • No validation
14. Phase 2: Structured tool calling Providing ways to ask for help • De fi ned tool methods • Typed parameters • Schema enforcement • Deterministic execution
15. Demo: Tool calling Adds fl exibility with educated guidance
16. Shift of control Back into the hands of developers!
17. Act 3: Context is key
18. Why prompts aren’t enough -> context engineering • LLMs are holding it all • Prompt • Data • Previous highlights • Repetitive for each request! https://neo4j.com/blog/agentic-ai/hands-on-with-context-graphs-and-neo4j/
19. Retrieval strategies • Embedding-based retrieval • Structured database queries • Additional traversal (e.g. graph) • Conversation state
20. Memory is a system responsibility Doesn’t live in the model • LLMs are stateless • Every request is independent • Conversation history stored externally • Domain knowledge retrieved intentionally • Memory is infrastructure, not a prompt trick
21. Graph as application memory • Short-term memory • Conversation state • Long-term memory • Users <-> Preferences • Why graph? • Relationships are explicit • Context retrieval = deterministic and precise https://neo4j.com/blog/developer/meet-lennys-memory-building-context-graphs-for-ai-agents/
22. Demo: Conversation memory Adds context over time and repetition
23. Act 4: Decoupling
24. Agent scaling problem • Initial agent construction: • Manual tool wiring • Model-speci fi c integrations • Tight coupling • Hard to swap providers
25. The solution…. Model Context Protocol (MCP) • Open protocol • Provides: • Tool discovery • Capability descriptions • Structured invocation • Standard responses • Universal adapter for AI
26. MCP architecture
27. Architecture with MCP + Neo4j
28. Demo: Tool discovery Adds modularity and agility
29. Act 5: Tying it together
30. Four layers of modern AI systems • 🧠 - LLM reasoning • ✋ - tool execution • 🗂 - graph-powered context • 🔌 - MCP standardization
31. Resources • Code: github.com/JMHReif/agents-mcp-ohmy • Come visit us at the booth! Jennifer Reif jennifer.reif@neo4j.com @JMHReif github.com/JMHReif jmhreif.com linkedin.com/in/jmhreif/