Need standardized way to specify model requirements during sampling #20

Currently, the Model Context Protocol doesn't provide a way for servers to indicate model requirements when making sampling requests. This creates several challenges:

Current Limitations

1. No way to specify minimum capability requirements
2. No way to specify cost constraints
3. No way to indicate preferred models
4. No standardized model naming across providers (e.g., gpt-4 vs claude-2)

Impact

This limitation means that:

• Servers can't ensure their prompts will work with the selected model
• Cost management becomes difficult
• Can't optimize for specific model strengths (e.g haiku vs sonnet)
• No guarantee of consistent behavior across different clients