Clarifying AI Concepts: From LLMs and MCP to Context and Statelessness
4310sy · x · 2026-08-11
This article systematically clarifies commonly confused AI concepts, helping developers understand AI tools from an underlying mechanism perspective.
- Model vs. Application: An LLM is essentially a mathematical formula predicting the next token. Actions like modifying files or executing commands are handled by the outer application layer (e.g., Cursor).
- API Statelessness: LLMs do not retain session state. The application must resend the full conversation history with each request, and excessively large contexts can trigger the 'Lost in the Middle' problem due to attention dispersion.
- Role of MCP: A standardized communication protocol based on JSON-RPC 2.0 that connects AI applications to external data sources, providing capabilities like Tools and Resources.
More from coding & agent
- Docker Is Isolation, Not Enforcement: Why Containers Aren't Sandboxes — max_paperclips · 2026-08-11
- Awesome A2A: A Curated GitHub List for the Agent2Agent Protocol Ecosystem — tom_doerr · 2026-08-11
- Stop Pre-injecting Context: Let Agents Search to Reduce Errors — siddharthnibjiya · 2026-08-11
- Orchestrating 12 Claude Agents to Write a Paper and Build a Website — sebkrier · 2026-08-11
- Opus 5 + Higgsfield One-Shot Generates 3D Parkour Game with Zero Hand-Modeled Assets — Vjeux · 2026-08-11
- Coinbase, Shopify, Ramp Build Custom Coding Agents but Still Pay Anthropic — Familiar_Jaguar_4134 · 2026-08-11