LLMs, Meet Your Tools: The Magic of Model Context Protocol
summary
Model Context Protocol (MCP) is a JSON‑RPC 2.0 based standard introduced Dec 2024 and already adopted by major providers, letting LLMs call JSON-RPC 2.0 servers to run tools, fetch resources, use prompt templates and (rarely) sampling. Demos use a VS Code client (Zenoder), playwright MCP to drive browsers, and GitHub/Postgres/Grafana MCPs; a prompt‑injection exploit silently creates local files, showing tool shadowing and tool‑poisoning risks. Defenses: whitelist MCPs, org configs, readonly flags, OAuth and third‑party prompt filters.
AI-written summary · about
chapters (12)
- 0:00Introduction and speaker background
- 1:49LLMs limitations and lack of grounding
- 4:39Introduction to Model Context Protocol
- 6:26MCP server and host explained
- 8:12MCP features beyond tools
- 13:37MCP tools and security concerns
- 17:35Practical MCP examples
- 23:18Creating custom MCP servers
- 25:07Security risks in MCP usage
- 28:22Live demo of prompt injection attacks
- 33:46Mitigating MCP security vulnerabilities
- 37:23Q&A and closing remarks
related talks
34mLLMs, Meet Your Tools: The Magic of Model Context Protocol
32mMCP for dummies: Feed more context to LLM
46mMCP for LLM Agents, Infrastructure APIs & Data Services
44mSupercharging DevOps with MCP (Without Opening a Security Hole)
50mBeyond local tools: Dive into the Model Context Protocol (MCP)
44mBeyond the Prompt: How MCP Connects Language Models to Real Tools
33mThe Model Context Protocol: Connecting AI to Everything
32m30 minutes to understand MCP (Model Context Protocol)