← back
Full Spec MCP: Hidden Capabilities of the MCP spec — Harald Kirschner, Microsoft/VSCode
Takeaway
Full MCP spec — sampling, resources, dynamic discovery, roots — unlocks rich stateful agent UX far beyond the tool-only patterns most servers ship today.
Summary
- VS Code shipping full MCP spec support in v1.10 (insiders): tools, resources, prompts, sampling, roots, plus dynamic discovery.
- Most servers stop at tools ('MCP as API wrapper syndrome'); LangChain research shows agents fail with too many tools, too many domains, or repetition — quality over quantity.
- VS Code adds per-chat tool selection (toolpacker), tool mentions, and user-defined tool sets for reusable workflows like 'frontend testing'.
- Sampling primitive (server requests LLM completions from client) is now implementable; enables agentic servers running LLM calls via the client (wired to GPT-4.1).
- Dynamic discovery demoed via a MUD-MCP dungeon game where the 'battle' tool only appears when a monster is in the room.
mcpvscodetools
Original description
The true power of Model Context Protocol emerges when clients and servers collaborate across the full spectrum of the specification. This talk presents practical examples of how VS Code's comprehensive implementation of MCP transforms the capabilities of AI assistants, making them more contextual, efficient, and user-friendly. We'll showcase advanced features like dynamic tool discovery and workspace-aware roots, demonstrating how they create experiences impossible with standard tools integrations while confronting the reality gap between MCP's theoretical potential and practical implementation challenges. Timestamps [00:00] Introduction to MCP and its Current State [01:42] The "MCP is just another API wrapper" Syndrome [02:51] VS Code's Full Spec Support [03:40] Challenges with Tools and Solutions [06:35] Resources and Their Importance [07:41] Sampling [09:21] Developer Experience Improvements [10:28] Staying Updated with the Spec [10:53] Key Upcoming Features and Community Efforts [12:46] Call to Action