Qdrant RAG MCP Server

# Progressive Context Integration Strategy ## Decision: Enhance Existing Search Tools After analyzing our codebase, we'll integrate Progressive Context Management into our existing search tools rather than creating separate ones. ## Integration Design ### 1. Enhanced Parameters for Existing Tools Add these parameters to `search()`, `search_code()`, and `search_docs()`: ```python def search( query: str, n_results: int = 5, # Existing parameters... search_mode: str = "hybrid", include_context: bool = True, context_chunks: int = 1, include_dependencies: bool = False, # New progressive context parameters context_level: str = "auto", # "auto", "file", "class", "method", "full" progressive_mode: bool = None, # None = auto based on context_level include_expansion_options: bool = True, semantic_cache: bool = True ) -> Dict[str, Any]: ``` ### 2. Context Levels - **"auto"** (default): Use query intent classification to determine level - **"file"**: Return file-level summaries (70% token reduction) - **"class"**: Return class/function signatures (50% token reduction) - **"method"**: Return full implementations (20% token reduction) - **"full"**: Current behavior (no reduction) ### 3. Backward Compatibility ```python # Old usage still works exactly the same search(query="authentication", n_results=5) # New progressive usage search(query="authentication", context_level="file") # Explicit progressive mode search(query="authentication", progressive_mode=True) ``` ### 4. Response Structure Enhancement Current response structure remains, with optional progressive fields: ```python { "results": [...], # Same as before "query": "...", "total": 10, # New progressive fields (only when progressive_mode=True) "progressive": { "level_used": "file", "token_estimate": 1500, "token_reduction": "70%", "expansion_options": [ { "type": "class", "path": "auth/authenticator.py::Authenticator", "estimated_tokens": 800, "relevance": 0.92 } ], "cache_hit": true, "query_intent": { "type": "exploration", "confidence": 0.85 } } } ``` ### 5. Implementation Strategy #### Phase 1: Infrastructure (Behind Feature Flag) ```python # In server_config.json "progressive_context": { "enabled": false, # Feature flag "default_level": "auto", ... } # In search functions if config.get("progressive_context", {}).get("enabled", False): # Use progressive manager else: # Current implementation ``` #### Phase 2: Gradual Rollout 1. Start with `search()` function only 2. Add to `search_code()` after validation 3. Finally add to `search_docs()` #### Phase 3: Make Default Once proven stable, flip the default to enabled ### 6. Query Intent Auto-Detection When `context_level="auto"` (default), use patterns: ```python # High-level exploration → file level "what does X do", "explain", "overview", "architecture" # Debugging/implementation → method level "bug in", "error", "line", "implementation of" # Navigation → class level "find", "where is", "show me" ``` ### 7. Synergy with Existing Features ```python # Progressive + context expansion search(query="auth", context_level="class", include_context=True) # Returns: Class signatures + surrounding file context # Progressive + dependencies search_code(query="login", context_level="file", include_dependencies=True) # Returns: File summaries + imported file summaries # Progressive + search modes search(query="config", context_level="file", search_mode="keyword") # Returns: File summaries from keyword search ``` ## Benefits of This Approach 1. **No Learning Curve**: Users continue using familiar tools 2. **Gradual Adoption**: Can enable progressively with feature flags 3. **Backward Compatible**: Existing code/scripts continue working 4. **Natural Evolution**: Follows established enhancement pattern 5. **Unified Experience**: One set of tools for all search needs ## Code Architecture ``` src/ ├── utils/ │ ├── progressive_context.py # New module │ │ ├── ProgressiveContextManager │ │ ├── SemanticCache │ │ ├── HierarchyBuilder │ │ └── QueryIntentClassifier │ └── ...existing utils... └── qdrant_mcp_context_aware.py ├── search() # Enhanced with progressive ├── search_code() # Enhanced with progressive └── search_docs() # Enhanced with progressive ``` ## Migration Path 1. **v0.3.2-alpha**: Feature flag off, infrastructure in place 2. **v0.3.2-beta**: Feature flag on for early adopters 3. **v0.3.2**: Feature enabled by default, can be disabled 4. **v0.4.0**: Remove feature flag, fully integrated This integration strategy provides the best balance of innovation and stability, allowing us to deliver Progressive Context Management without disrupting existing users.