---
name: MCP Sampling Support
about: Add MCP sampling capability for enhanced LLM-powered responses
title: '[FEATURE] Add MCP Sampling for Dynamic Content Generation'
labels: enhancement, mcp-protocol
assignees: ''
---
## Feature Request: MCP Sampling Support
### Description
Enable MCP sampling to allow the server to request LLM-generated content for enhanced, context-aware responses.
### Use Cases
#### 1. Dynamic Swiss German Explanations
Instead of hardcoded translations, generate contextual, cultural explanations:
```python
explanation = await mcp.request_sampling(
messages=[{
"role": "user",
"content": f"Explain the Swiss German phrase '{swiss_german_text}' in context of swimming in Aare river"
}]
)
```
#### 2. Intelligent Safety Warnings
Generate context-aware safety messages based on multiple factors:
```python
safety_warning = await mcp.request_sampling(
messages=[{
"role": "user",
"content": f"Generate a safety warning for: flow={flow}m³/s, temp={temp}°C, weather={weather}. Be clear and direct."
}]
)
```
#### 3. Historical Trend Summaries
Let the LLM summarize complex trend data:
```python
trend_summary = await mcp.request_sampling(
messages=[{
"role": "user",
"content": f"Summarize this week's water temperature trend: {historical_data}. Highlight key patterns."
}]
)
```
#### 4. Comparative City Analysis
Enhanced multi-city comparisons with natural language:
```python
comparison = await mcp.request_sampling(
messages=[{
"role": "user",
"content": f"Compare these swimming locations and recommend the best: {cities_data}"
}]
)
```
### Technical Considerations
**Pros:**
- ✅ More natural, context-aware responses
- ✅ Dynamic content generation
- ✅ Better personalization
- ✅ Reduced hardcoded text
- ✅ Cultural context for Swiss German phrases
**Cons:**
- ❌ Adds latency (LLM call per sampling request)
- ❌ Requires client sampling support
- ❌ Less predictable/testable
- ❌ May duplicate work (client would call LLM anyway)
### Implementation Plan
1. **Check FastMCP 2.0 Support**
- Verify if `mcp.request_sampling()` is available
- Check MCP protocol version requirements
2. **Add Sampling Helpers**
```python
async def generate_safety_warning(flow: float, temp: float) -> str:
if not hasattr(mcp, 'request_sampling'):
return _static_safety_warning(flow, temp)
return await mcp.request_sampling(
messages=[...],
max_tokens=150
)
```
3. **Optional Enhancement**
- Keep current simple responses as default
- Add `--enable-sampling` flag or config option
- Use sampling only when it adds clear value
4. **Priority Areas**
- Safety warnings (critical, context-dependent)
- Swiss German cultural explanations (valuable context)
- Comparative analyses (complex multi-city comparisons)
### Configuration
Add settings for sampling control:
```python
class Settings(BaseSettings):
# Sampling configuration
enable_sampling: bool = Field(
default=False,
description="Enable MCP sampling for enhanced responses"
)
sampling_max_tokens: int = Field(
default=150,
description="Maximum tokens for sampling requests"
)
sampling_timeout: float = Field(
default=10.0,
description="Timeout for sampling requests in seconds"
)
```
### Testing Requirements
- [ ] Test sampling availability detection
- [ ] Test fallback when sampling unavailable
- [ ] Test sampling timeout handling
- [ ] Test sampling error handling
- [ ] Test response quality with/without sampling
- [ ] Performance benchmarks with sampling enabled
### Dependencies
- FastMCP 2.0+ with sampling support
- MCP protocol version with sampling capability
- Client support for sampling (Claude Desktop, etc.)
### Related
- MCP Protocol Specification: Sampling
- FastMCP Documentation
- Issue #XXX: Improve Swiss German translations
### Priority
**Medium** - Enhancement that improves user experience but not critical for core functionality.
Start with:
1. Swiss German cultural explanations (high value, low latency impact)
2. Safety warnings (critical content, worth the latency)
3. Historical trend summaries (complex analysis benefit)
