claude-self-reflect 3.2.4 → 3.3.0

package/mcp-server/src/server.py

@@ -4,9 +4,8 @@ import os
 import asyncio
 from pathlib import Path
 from typing import Any, Optional, List, Dict, Union
-from datetime import datetime, timezone
+from datetime import datetime, timezone, timedelta
 import json
-import numpy as np
 import hashlib
 import time
 import logging
@@ -27,10 +26,14 @@ except ImportError:
     logging.warning("Using legacy utils.normalize_project_name - shared module not found")
 
 from .project_resolver import ProjectResolver
+from .temporal_utils import SessionDetector, TemporalParser, WorkSession, group_by_time_period
+from .temporal_tools import register_temporal_tools
+from .search_tools import register_search_tools
+from .reflection_tools import register_reflection_tools
 from pydantic import BaseModel, Field
 from qdrant_client import AsyncQdrantClient, models
 from qdrant_client.models import (
-    PointStruct, VectorParams, Distance
+    PointStruct, VectorParams, Distance, OrderBy
 )
 
 # Try to import newer Qdrant API for native decay
@@ -85,11 +88,49 @@ EMBEDDING_MODEL = os.getenv('EMBEDDING_MODEL', 'sentence-transformers/all-MiniLM
 # Import the robust embedding manager
 from .embedding_manager import get_embedding_manager
 
+# Import new performance modules with fallback
+try:
+    from .connection_pool import QdrantConnectionPool, CircuitBreaker
+    CONNECTION_POOL_AVAILABLE = True
+except ImportError:
+    CONNECTION_POOL_AVAILABLE = False
+    logging.warning("Connection pool module not available")
+
+try:
+    from .parallel_search import parallel_search_collections
+    PARALLEL_SEARCH_AVAILABLE = True
+except ImportError:
+    PARALLEL_SEARCH_AVAILABLE = False
+    logging.warning("Parallel search module not available")
+
+# Set default configuration values
+MAX_RESULTS_PER_COLLECTION = 10
+MAX_TOTAL_RESULTS = 1000
+MAX_MEMORY_MB = 500
+POOL_SIZE = 10
+POOL_MAX_OVERFLOW = 5
+POOL_TIMEOUT = 30.0
+RETRY_ATTEMPTS = 3
+RETRY_DELAY = 1.0
+MAX_CONCURRENT_SEARCHES = 10
+ENABLE_PARALLEL_SEARCH = True
+
+try:
+    from .decay_manager import DecayManager
+    DECAY_MANAGER_AVAILABLE = True
+except ImportError:
+    DECAY_MANAGER_AVAILABLE = False
+    logging.warning("Decay manager module not available")
+
 # Lazy initialization - models will be loaded on first use
 embedding_manager = None
 voyage_client = None  # Keep for backward compatibility
 local_embedding_model = None  # Keep for backward compatibility
 
+# Initialize connection pool
+qdrant_pool = None
+circuit_breaker = None
+
 def initialize_embeddings():
     """Initialize embedding models with robust fallback."""
     global embedding_manager, voyage_client, local_embedding_model
@@ -101,7 +142,7 @@ def initialize_embeddings():
         if embedding_manager.model_type == 'voyage':
             voyage_client = embedding_manager.voyage_client
         elif embedding_manager.model_type == 'local':
-            local_embedding_model = embedding_manager.model
+            local_embedding_model = embedding_manager.local_model
         
         return True
     except Exception as e:
@@ -109,9 +150,8 @@ def initialize_embeddings():
         return False
 
 # Debug environment loading and startup
-import sys
-import datetime as dt
-startup_time = dt.datetime.now().isoformat()
+# Debug environment loading and startup
+startup_time = datetime.now(timezone.utc).isoformat()
 print(f"[STARTUP] MCP Server starting at {startup_time}", file=sys.stderr)
 print(f"[STARTUP] Python: {sys.version}", file=sys.stderr)
 print(f"[STARTUP] Working directory: {os.getcwd()}", file=sys.stderr)
@@ -152,8 +192,32 @@ mcp = FastMCP(
     instructions="Search past conversations and store reflections with time-based memory decay"
 )
 
-# Create Qdrant client
-qdrant_client = AsyncQdrantClient(url=QDRANT_URL)
+# Initialize Qdrant client with connection pooling if available
+if CONNECTION_POOL_AVAILABLE and ENABLE_PARALLEL_SEARCH:
+    qdrant_pool = QdrantConnectionPool(
+        url=QDRANT_URL,
+        pool_size=POOL_SIZE,
+        max_overflow=POOL_MAX_OVERFLOW,
+        timeout=POOL_TIMEOUT,
+        retry_attempts=RETRY_ATTEMPTS,
+        retry_delay=RETRY_DELAY
+    )
+    # Create a wrapper for backward compatibility
+    qdrant_client = AsyncQdrantClient(url=QDRANT_URL)
+    circuit_breaker = CircuitBreaker(failure_threshold=5, recovery_timeout=60.0)
+    print(f"[INFO] Connection pool initialized with size {POOL_SIZE}", file=sys.stderr)
+else:
+    # Fallback to single client
+    qdrant_client = AsyncQdrantClient(url=QDRANT_URL)
+    qdrant_pool = None
+    circuit_breaker = None
+    print(f"[INFO] Using single Qdrant client (no pooling)", file=sys.stderr)
+
+# Initialize decay manager if available
+decay_manager = None
+if DECAY_MANAGER_AVAILABLE:
+    decay_manager = DecayManager()
+    print(f"[INFO] Decay manager initialized", file=sys.stderr)
 
 # Add MCP Resources for system status
 @mcp.resource("status://import-stats")
@@ -492,7 +556,14 @@ def get_embedding_dimension() -> int:
 
 def get_collection_suffix() -> str:
     """Get the collection suffix based on embedding provider."""
-    if PREFER_LOCAL_EMBEDDINGS or not voyage_client:
+    # Use embedding_manager's model type if available
+    if embedding_manager and hasattr(embedding_manager, 'model_type'):
+        if embedding_manager.model_type == 'voyage':
+            return "_voyage"
+        else:
+            return "_local"
+    # Fallback to environment variable
+    elif PREFER_LOCAL_EMBEDDINGS:
         return "_local"
     else:
         return "_voyage"
@@ -578,1718 +649,54 @@ def aggregate_pattern_intelligence(results: List[SearchResult]) -> Dict[str, Any
     return intelligence
     
 # Register tools
-@mcp.tool()
-async def reflect_on_past(
-    ctx: Context,
-    query: str = Field(description="The search query to find semantically similar conversations"),
-    limit: int = Field(default=5, description="Maximum number of results to return"),
-    min_score: float = Field(default=0.3, description="Minimum similarity score (0-1)"),
-    use_decay: Union[int, str] = Field(default=-1, description="Apply time-based decay: 1=enable, 0=disable, -1=use environment default (accepts int or str)"),
-    project: Optional[str] = Field(default=None, description="Search specific project only. If not provided, searches current project based on working directory. Use 'all' to search across all projects."),
-    include_raw: bool = Field(default=False, description="Include raw Qdrant payload data for debugging (increases response size)"),
-    response_format: str = Field(default="xml", description="Response format: 'xml' or 'markdown'"),
-    brief: bool = Field(default=False, description="Brief mode: returns minimal information for faster response"),
-    mode: str = Field(default="full", description="Search mode: 'full' (all results with details), 'quick' (count + top result only), 'summary' (aggregated insights without individual results)")
-) -> str:
-    """Search for relevant past conversations using semantic search with optional time decay."""
-    
-    logger.info(f"=== SEARCH START === Query: '{query}', Project: '{project}', Limit: {limit}")
-    
-    # Validate mode parameter
-    if mode not in ['full', 'quick', 'summary']:
-        return f"<error>Invalid mode '{mode}'. Must be 'full', 'quick', or 'summary'</error>"
-    
-    # Start timing
-    start_time = time.time()
-    timing_info = {}
-    
-    # Normalize use_decay to integer
-    timing_info['param_parsing_start'] = time.time()
-    if isinstance(use_decay, str):
-        try:
-            use_decay = int(use_decay)
-        except ValueError:
-            raise ValueError("use_decay must be '1', '0', or '-1'")
-    timing_info['param_parsing_end'] = time.time()
-    
-    # Parse decay parameter using integer approach
-    should_use_decay = (
-        True if use_decay == 1
-        else False if use_decay == 0
-        else ENABLE_MEMORY_DECAY  # -1 or any other value
-    )
-    
-    # Determine project scope
-    target_project = project
-    
-    # Always get the working directory for logging purposes
-    cwd = os.environ.get('MCP_CLIENT_CWD', os.getcwd())
-    await ctx.debug(f"CWD: {cwd}, Project param: {project}")
-    
-    if project is None:
-        # Use MCP_CLIENT_CWD environment variable set by run-mcp.sh
-        # This contains the actual working directory where Claude Code is running
-        
-        # Extract project name from path (e.g., /Users/.../projects/project-name)
-        path_parts = Path(cwd).parts
-        if 'projects' in path_parts:
-            idx = path_parts.index('projects')
-            if idx + 1 < len(path_parts):
-                target_project = path_parts[idx + 1]
-        elif '.claude' in path_parts:
-            # If we're in a .claude directory, go up to find project
-            for i, part in enumerate(path_parts):
-                if part == '.claude' and i > 0:
-                    target_project = path_parts[i - 1]
-                    break
-        
-        # If still no project detected, use the last directory name
-        if target_project is None:
-            target_project = Path(cwd).name
-        
-        await ctx.debug(f"Auto-detected project from path: {target_project}")
-    
-    # For project matching, we need to handle the dash-encoded format
-    # Convert folder name to the format used in stored data
-    if target_project != 'all':
-        # The stored format uses full path with dashes, so we need to construct it
-        # For now, let's try to match based on the end of the project name
-        pass  # We'll handle this differently in the filtering logic
-    
-    await ctx.debug(f"Searching for: {query}")
-    await ctx.debug(f"Client working directory: {cwd}")
-    await ctx.debug(f"Project scope: {target_project if target_project != 'all' else 'all projects'}")
-    await ctx.debug(f"Decay enabled: {should_use_decay}")
-    await ctx.debug(f"Native decay mode: {USE_NATIVE_DECAY}")
-    await ctx.debug(f"ENABLE_MEMORY_DECAY env: {ENABLE_MEMORY_DECAY}")
-    await ctx.debug(f"DECAY_WEIGHT: {DECAY_WEIGHT}, DECAY_SCALE_DAYS: {DECAY_SCALE_DAYS}")
-    
-    try:
-        # We'll generate embeddings on-demand per collection type
-        timing_info['embedding_prep_start'] = time.time()
-        query_embeddings = {}  # Cache embeddings by type
-        timing_info['embedding_prep_end'] = time.time()
-        
-        # Get all collections
-        timing_info['get_collections_start'] = time.time()
-        all_collections = await get_all_collections()
-        timing_info['get_collections_end'] = time.time()
-        
-        if not all_collections:
-            return "No conversation collections found. Please import conversations first."
-        
-        # Filter collections by project if not searching all
-        project_collections = []  # Define at this scope for later use
-        if target_project != 'all':
-            # Use ProjectResolver with sync client (resolver expects sync operations)
-            from qdrant_client import QdrantClient as SyncQdrantClient
-            sync_client = SyncQdrantClient(url=QDRANT_URL)
-            resolver = ProjectResolver(sync_client)
-            project_collections = resolver.find_collections_for_project(target_project)
-            
-            if not project_collections:
-                # Fall back to old method for backward compatibility
-                normalized_name = normalize_project_name(target_project)
-                project_hash = hashlib.md5(normalized_name.encode()).hexdigest()[:8]
-                project_collections = [
-                    c for c in all_collections 
-                    if c.startswith(f"conv_{project_hash}_")
-                ]
-            
-            # Always include reflections collections when searching a specific project
-            reflections_collections = [c for c in all_collections if c.startswith('reflections')]
-            
-            if not project_collections:
-                # Fall back to searching all collections but filtering by project metadata
-                await ctx.debug(f"No collections found for project {target_project}, will filter by metadata")
-                collections_to_search = all_collections
-            else:
-                await ctx.debug(f"Found {len(project_collections)} collections for project {target_project}")
-                # Include both project collections and reflections
-                collections_to_search = project_collections + reflections_collections
-                # Remove duplicates
-                collections_to_search = list(set(collections_to_search))
-        else:
-            collections_to_search = all_collections
-        
-        await ctx.debug(f"Searching across {len(collections_to_search)} collections")
-        await ctx.debug(f"Using {'local' if PREFER_LOCAL_EMBEDDINGS or not voyage_client else 'Voyage AI'} embeddings")
-        
-        all_results = []
-        
-        # Search each collection
-        timing_info['search_all_start'] = time.time()
-        collection_timings = []
-        
-        # Report initial progress
-        await ctx.report_progress(progress=0, total=len(collections_to_search))
-        
-        for idx, collection_name in enumerate(collections_to_search):
-            collection_timing = {'name': collection_name, 'start': time.time()}
-            
-            # Report progress before searching each collection
-            await ctx.report_progress(
-                progress=idx, 
-                total=len(collections_to_search),
-                message=f"Searching {collection_name}"
-            )
-            
-            try:
-                # Determine embedding type for this collection
-                embedding_type_for_collection = 'voyage' if collection_name.endswith('_voyage') else 'local'
-                
-                # Generate or retrieve cached embedding for this type
-                if embedding_type_for_collection not in query_embeddings:
-                    try:
-                        query_embeddings[embedding_type_for_collection] = await generate_embedding(query, force_type=embedding_type_for_collection)
-                    except Exception as e:
-                        await ctx.debug(f"Failed to generate {embedding_type_for_collection} embedding: {e}")
-                        continue
-                
-                query_embedding = query_embeddings[embedding_type_for_collection]
-                
-                if should_use_decay and USE_NATIVE_DECAY and NATIVE_DECAY_AVAILABLE:
-                    # Use native Qdrant decay with newer API
-                    await ctx.debug(f"Using NATIVE Qdrant decay (new API) for {collection_name}")
-                    
-                    # Build the query with native Qdrant decay formula using newer API
-                    # Convert half-life to seconds (Qdrant uses seconds for datetime)
-                    half_life_seconds = DECAY_SCALE_DAYS * 24 * 60 * 60
-                    
-                    # Build query using proper Python models as per Qdrant docs
-                    from qdrant_client import models
-                    
-                    query_obj = models.FormulaQuery(
-                        formula=models.SumExpression(
-                            sum=[
-                                "$score",  # Original similarity score
-                                models.MultExpression(
-                                    mult=[
-                                        DECAY_WEIGHT,  # Weight multiplier
-                                        models.ExpDecayExpression(
-                                            exp_decay=models.DecayParamsExpression(
-                                                x=models.DatetimeKeyExpression(
-                                                    datetime_key="timestamp"  # Payload field with datetime
-                                                ),
-                                                target=models.DatetimeExpression(
-                                                    datetime="now"  # Current time on server
-                                                ),
-                                                scale=half_life_seconds,  # Scale in seconds
-                                                midpoint=0.5  # Half-life semantics
-                                            )
-                                        )
-                                    ]
-                                )
-                            ]
-                        )
-                    )
-                    
-                    # Execute query with native decay (new API)
-                    results = await qdrant_client.query_points(
-                        collection_name=collection_name,
-                        query=query_obj,
-                        limit=limit,
-                        with_payload=True
-                        # No score_threshold - let Qdrant's decay formula handle relevance
-                    )
-                elif should_use_decay and USE_NATIVE_DECAY and not NATIVE_DECAY_AVAILABLE:
-                    # Use native Qdrant decay with older API
-                    await ctx.debug(f"Using NATIVE Qdrant decay (legacy API) for {collection_name}")
-                    
-                    # Convert half-life to seconds (Qdrant uses seconds for datetime)
-                    half_life_seconds = DECAY_SCALE_DAYS * 24 * 60 * 60
-                    
-                    # Build the query with native Qdrant decay formula using older API
-                    # Use the same models but with FormulaQuery
-                    query_obj = FormulaQuery(
-                        nearest=query_embedding,
-                        formula=SumExpression(
-                            sum=[
-                                "$score",  # Original similarity score
-                                {
-                                    "mult": [
-                                        DECAY_WEIGHT,  # Weight multiplier
-                                        {
-                                            "exp_decay": DecayParamsExpression(
-                                                x=DatetimeKeyExpression(datetime_key="timestamp"),
-                                                target=DatetimeExpression(datetime="now"),
-                                                scale=half_life_seconds,  # Scale in seconds
-                                                midpoint=0.5  # Half-life semantics
-                                            )
-                                        }
-                                    ]
-                                }
-                            ]
-                        )
-                    )
-                    
-                    # Execute query with native decay
-                    results = await qdrant_client.query_points(
-                        collection_name=collection_name,
-                        query=query_obj,
-                        limit=limit,
-                        with_payload=True
-                        # No score_threshold - let Qdrant's decay formula handle relevance
-                    )
-                    
-                    # Process results from native decay search
-                    for point in results.points:
-                        # Clean timestamp for proper parsing
-                        raw_timestamp = point.payload.get('timestamp', datetime.now().isoformat())
-                        clean_timestamp = raw_timestamp.replace('Z', '+00:00') if raw_timestamp.endswith('Z') else raw_timestamp
-                        
-                        # Check project filter if we're searching all collections but want specific project
-                        point_project = point.payload.get('project', collection_name.replace('conv_', '').replace('_voyage', '').replace('_local', ''))
-                        
-                        # Special handling for reflections - they're global by default but can have project context
-                        is_reflection_collection = collection_name.startswith('reflections')
-                        
-                        # Handle project matching - check if the target project name appears at the end of the stored project path
-                        if target_project != 'all' and not project_collections and not is_reflection_collection:
-                            # The stored project name is like "-Users-username-projects-ShopifyMCPMockShop"
-                            # We want to match just "ShopifyMCPMockShop"
-                            # Also handle underscore/dash variations (procsolve-website vs procsolve_website)
-                            normalized_target = target_project.replace('-', '_')
-                            normalized_stored = point_project.replace('-', '_')
-                            if not (normalized_stored.endswith(f"_{normalized_target}") or 
-                                    normalized_stored == normalized_target or
-                                    point_project.endswith(f"-{target_project}") or 
-                                    point_project == target_project):
-                                continue  # Skip results from other projects
-                        
-                        # For reflections with project context, optionally filter by project
-                        if is_reflection_collection and target_project != 'all' and 'project' in point.payload:
-                            # Only filter if the reflection has project metadata
-                            reflection_project = point.payload.get('project', '')
-                            if reflection_project:
-                                # Normalize both for comparison (handle underscore/dash variations)
-                                normalized_target = target_project.replace('-', '_')
-                                normalized_reflection = reflection_project.replace('-', '_')
-                                if not (
-                                    reflection_project == target_project or 
-                                    normalized_reflection == normalized_target or
-                                    reflection_project.endswith(f"/{target_project}") or
-                                    reflection_project.endswith(f"-{target_project}") or
-                                    normalized_reflection.endswith(f"_{normalized_target}") or
-                                    normalized_reflection.endswith(f"/{normalized_target}")
-                                ):
-                                    continue  # Skip reflections from other projects
-                            
-                        # Log pattern data
-                        patterns = point.payload.get('code_patterns')
-                        logger.info(f"DEBUG: Creating SearchResult for point {point.id} from {collection_name}: has_patterns={bool(patterns)}, pattern_keys={list(patterns.keys()) if patterns else None}")
-                        
-                        all_results.append(SearchResult(
-                            id=str(point.id),
-                            score=point.score,  # Score already includes decay
-                            timestamp=clean_timestamp,
-                            role=point.payload.get('start_role', point.payload.get('role', 'unknown')),
-                            excerpt=(point.payload.get('text', '')[:350] + '...' if len(point.payload.get('text', '')) > 350 else point.payload.get('text', '')),
-                            project_name=point_project,
-                            conversation_id=point.payload.get('conversation_id'),
-                            base_conversation_id=point.payload.get('base_conversation_id'),
-                            collection_name=collection_name,
-                            raw_payload=point.payload,  # Always include payload for metadata extraction
-                            # Pattern intelligence metadata
-                            code_patterns=point.payload.get('code_patterns'),
-                            files_analyzed=point.payload.get('files_analyzed'),
-                            tools_used=list(point.payload.get('tools_used', [])) if isinstance(point.payload.get('tools_used'), set) else point.payload.get('tools_used'),
-                            concepts=point.payload.get('concepts')
-                        ))
-                    
-                elif should_use_decay:
-                    # Use client-side decay (existing implementation)
-                    await ctx.debug(f"Using CLIENT-SIDE decay for {collection_name}")
-                    
-                    # Search without score threshold to get all candidates
-                    results = await qdrant_client.search(
-                        collection_name=collection_name,
-                        query_vector=query_embedding,
-                        limit=limit * 3,  # Get more candidates for decay filtering
-                        with_payload=True
-                    )
-                    
-                    # Apply decay scoring manually
-                    now = datetime.now(timezone.utc)
-                    scale_ms = DECAY_SCALE_DAYS * 24 * 60 * 60 * 1000
-                    
-                    decay_results = []
-                    for point in results:
-                        try:
-                            # Get timestamp from payload
-                            timestamp_str = point.payload.get('timestamp')
-                            if timestamp_str:
-                                timestamp = datetime.fromisoformat(timestamp_str.replace('Z', '+00:00'))
-                                # Ensure timestamp is timezone-aware
-                                if timestamp.tzinfo is None:
-                                    timestamp = timestamp.replace(tzinfo=timezone.utc)
-                                age_ms = (now - timestamp).total_seconds() * 1000
-                                
-                                # Calculate decay factor using proper half-life formula
-                                # For half-life H: decay = exp(-ln(2) * age / H)
-                                ln2 = math.log(2)
-                                decay_factor = math.exp(-ln2 * age_ms / scale_ms)
-                                
-                                # Apply multiplicative decay formula to keep scores bounded [0, 1]
-                                # adjusted = score * ((1 - weight) + weight * decay)
-                                adjusted_score = point.score * ((1 - DECAY_WEIGHT) + DECAY_WEIGHT * decay_factor)
-                                
-                                # Debug: show the calculation
-                                age_days = age_ms / (24 * 60 * 60 * 1000)
-                                await ctx.debug(f"Point: age={age_days:.1f} days, original_score={point.score:.3f}, decay_factor={decay_factor:.3f}, adjusted_score={adjusted_score:.3f}")
-                            else:
-                                adjusted_score = point.score
-                            
-                            # Only include if above min_score after decay
-                            if adjusted_score >= min_score:
-                                decay_results.append((adjusted_score, point))
-                        
-                        except Exception as e:
-                            await ctx.debug(f"Error applying decay to point: {e}")
-                            decay_results.append((point.score, point))
-                    
-                    # Sort by adjusted score and take top results
-                    decay_results.sort(key=lambda x: x[0], reverse=True)
-                    
-                    # Convert to SearchResult format
-                    for adjusted_score, point in decay_results[:limit]:
-                        # Clean timestamp for proper parsing
-                        raw_timestamp = point.payload.get('timestamp', datetime.now().isoformat())
-                        clean_timestamp = raw_timestamp.replace('Z', '+00:00') if raw_timestamp.endswith('Z') else raw_timestamp
-                        
-                        # Check project filter if we're searching all collections but want specific project
-                        point_project = point.payload.get('project', collection_name.replace('conv_', '').replace('_voyage', '').replace('_local', ''))
-                        
-                        # Special handling for reflections - they're global by default but can have project context
-                        is_reflection_collection = collection_name.startswith('reflections')
-                        
-                        # Handle project matching - check if the target project name appears at the end of the stored project path
-                        if target_project != 'all' and not project_collections and not is_reflection_collection:
-                            # The stored project name is like "-Users-username-projects-ShopifyMCPMockShop"
-                            # We want to match just "ShopifyMCPMockShop"
-                            # Also handle underscore/dash variations (procsolve-website vs procsolve_website)
-                            normalized_target = target_project.replace('-', '_')
-                            normalized_stored = point_project.replace('-', '_')
-                            if not (normalized_stored.endswith(f"_{normalized_target}") or 
-                                    normalized_stored == normalized_target or
-                                    point_project.endswith(f"-{target_project}") or 
-                                    point_project == target_project):
-                                continue  # Skip results from other projects
-                        
-                        # For reflections with project context, optionally filter by project
-                        if is_reflection_collection and target_project != 'all' and 'project' in point.payload:
-                            # Only filter if the reflection has project metadata
-                            reflection_project = point.payload.get('project', '')
-                            if reflection_project:
-                                # Normalize both for comparison (handle underscore/dash variations)
-                                normalized_target = target_project.replace('-', '_')
-                                normalized_reflection = reflection_project.replace('-', '_')
-                                if not (
-                                    reflection_project == target_project or 
-                                    normalized_reflection == normalized_target or
-                                    reflection_project.endswith(f"/{target_project}") or
-                                    reflection_project.endswith(f"-{target_project}") or
-                                    normalized_reflection.endswith(f"_{normalized_target}") or
-                                    normalized_reflection.endswith(f"/{normalized_target}")
-                                ):
-                                    continue  # Skip reflections from other projects
-                            
-                        all_results.append(SearchResult(
-                            id=str(point.id),
-                            score=adjusted_score,  # Use adjusted score
-                            timestamp=clean_timestamp,
-                            role=point.payload.get('start_role', point.payload.get('role', 'unknown')),
-                            excerpt=(point.payload.get('text', '')[:350] + '...' if len(point.payload.get('text', '')) > 350 else point.payload.get('text', '')),
-                            project_name=point_project,
-                            conversation_id=point.payload.get('conversation_id'),
-                            base_conversation_id=point.payload.get('base_conversation_id'),
-                            collection_name=collection_name,
-                            raw_payload=point.payload,  # Always include payload for metadata extraction
-                            # Pattern intelligence metadata
-                            code_patterns=point.payload.get('code_patterns'),
-                            files_analyzed=point.payload.get('files_analyzed'),
-                            tools_used=list(point.payload.get('tools_used', [])) if isinstance(point.payload.get('tools_used'), set) else point.payload.get('tools_used'),
-                            concepts=point.payload.get('concepts')
-                        ))
-                else:
-                    # Standard search without decay
-                    # Let Qdrant handle scoring natively
-                    results = await qdrant_client.search(
-                        collection_name=collection_name,
-                        query_vector=query_embedding,
-                        limit=limit * 2,  # Get more results to account for filtering
-                        with_payload=True
-                        # No score_threshold - let Qdrant decide what's relevant
-                    )
-                    
-                    for point in results:
-                        # Clean timestamp for proper parsing
-                        raw_timestamp = point.payload.get('timestamp', datetime.now().isoformat())
-                        clean_timestamp = raw_timestamp.replace('Z', '+00:00') if raw_timestamp.endswith('Z') else raw_timestamp
-                        
-                        # Check project filter if we're searching all collections but want specific project
-                        point_project = point.payload.get('project', collection_name.replace('conv_', '').replace('_voyage', '').replace('_local', ''))
-                        
-                        # Special handling for reflections - they're global by default but can have project context
-                        is_reflection_collection = collection_name.startswith('reflections')
-                        
-                        # Handle project matching - check if the target project name appears at the end of the stored project path
-                        if target_project != 'all' and not project_collections and not is_reflection_collection:
-                            # The stored project name is like "-Users-username-projects-ShopifyMCPMockShop"
-                            # We want to match just "ShopifyMCPMockShop"
-                            # Also handle underscore/dash variations (procsolve-website vs procsolve_website)
-                            normalized_target = target_project.replace('-', '_')
-                            normalized_stored = point_project.replace('-', '_')
-                            if not (normalized_stored.endswith(f"_{normalized_target}") or 
-                                    normalized_stored == normalized_target or
-                                    point_project.endswith(f"-{target_project}") or 
-                                    point_project == target_project):
-                                continue  # Skip results from other projects
-                        
-                        # For reflections with project context, optionally filter by project
-                        if is_reflection_collection and target_project != 'all' and 'project' in point.payload:
-                            # Only filter if the reflection has project metadata
-                            reflection_project = point.payload.get('project', '')
-                            if reflection_project:
-                                # Normalize both for comparison (handle underscore/dash variations)
-                                normalized_target = target_project.replace('-', '_')
-                                normalized_reflection = reflection_project.replace('-', '_')
-                                if not (
-                                    reflection_project == target_project or 
-                                    normalized_reflection == normalized_target or
-                                    reflection_project.endswith(f"/{target_project}") or
-                                    reflection_project.endswith(f"-{target_project}") or
-                                    normalized_reflection.endswith(f"_{normalized_target}") or
-                                    normalized_reflection.endswith(f"/{normalized_target}")
-                                ):
-                                    continue  # Skip reflections from other projects
-                        
-                        # BOOST V2 CHUNKS: Apply score boost for v2 chunks (better quality)
-                        original_score = point.score
-                        final_score = original_score
-                        chunking_version = point.payload.get('chunking_version', 'v1')
-                        
-                        if chunking_version == 'v2':
-                            # Boost v2 chunks by 20% (configurable)
-                            boost_factor = 1.2  # From migration config
-                            final_score = min(1.0, original_score * boost_factor)
-                            await ctx.debug(f"Boosted v2 chunk: {original_score:.3f} -> {final_score:.3f}")
-                        
-                        # Apply minimum score threshold after boosting
-                        if final_score < min_score:
-                            continue
-                            
-                        search_result = SearchResult(
-                            id=str(point.id),
-                            score=final_score,
-                            timestamp=clean_timestamp,
-                            role=point.payload.get('start_role', point.payload.get('role', 'unknown')),
-                            excerpt=(point.payload.get('text', '')[:350] + '...' if len(point.payload.get('text', '')) > 350 else point.payload.get('text', '')),
-                            project_name=point_project,
-                            conversation_id=point.payload.get('conversation_id'),
-                            base_conversation_id=point.payload.get('base_conversation_id'),
-                            collection_name=collection_name,
-                            raw_payload=point.payload,  # Always include payload for metadata extraction
-                            # Pattern intelligence metadata
-                            code_patterns=point.payload.get('code_patterns'),
-                            files_analyzed=point.payload.get('files_analyzed'),
-                            tools_used=list(point.payload.get('tools_used', [])) if isinstance(point.payload.get('tools_used'), set) else point.payload.get('tools_used'),
-                            concepts=point.payload.get('concepts')
-                        )
-                        
-                        all_results.append(search_result)
-            
-            except Exception as e:
-                await ctx.debug(f"Error searching {collection_name}: {str(e)}")
-                collection_timing['error'] = str(e)
-            
-            collection_timing['end'] = time.time()
-            collection_timings.append(collection_timing)
-        
-        timing_info['search_all_end'] = time.time()
-        
-        # Report completion of search phase
-        await ctx.report_progress(
-            progress=len(collections_to_search), 
-            total=len(collections_to_search),
-            message="Search complete, processing results"
-        )
-        
-        # Apply base_conversation_id boosting before sorting
-        timing_info['boost_start'] = time.time()
-        
-        # Group results by base_conversation_id to identify related chunks
-        base_conversation_groups = {}
-        for result in all_results:
-            base_id = result.base_conversation_id
-            if base_id:
-                if base_id not in base_conversation_groups:
-                    base_conversation_groups[base_id] = []
-                base_conversation_groups[base_id].append(result)
-        
-        # Apply boost to results from base conversations with multiple high-scoring chunks
-        base_conversation_boost = 0.1  # Boost factor for base conversation matching
-        for base_id, group_results in base_conversation_groups.items():
-            if len(group_results) > 1:  # Multiple chunks from same base conversation
-                avg_score = sum(r.score for r in group_results) / len(group_results)
-                if avg_score > 0.8:  # Only boost high-quality base conversations
-                    for result in group_results:
-                        result.score += base_conversation_boost
-                        await ctx.debug(f"Boosted result from base_conversation_id {base_id}: {result.score:.3f}")
-        
-        timing_info['boost_end'] = time.time()
-        
-        # Sort by score and limit
-        timing_info['sort_start'] = time.time()
-        all_results.sort(key=lambda x: x.score, reverse=True)
-        
-        # Apply mode-specific limits
-        if mode == "quick":
-            # For quick mode, only keep the top result
-            all_results = all_results[:1]
-        elif mode == "summary":
-            # For summary mode, we'll process all results but not return individual ones
-            pass  # Keep all for aggregation
-        else:
-            # For full mode, apply the normal limit
-            all_results = all_results[:limit]
-        
-        timing_info['sort_end'] = time.time()
-        
-        logger.info(f"Total results: {len(all_results)}, Mode: {mode}, Returning: {len(all_results[:limit])}")
-        for r in all_results[:3]:  # Log first 3
-            logger.debug(f"Result: id={r.id}, has_patterns={bool(r.code_patterns)}, pattern_keys={list(r.code_patterns.keys()) if r.code_patterns else None}")
-        
-        if not all_results:
-            return f"No conversations found matching '{query}'. Try different keywords or check if conversations have been imported."
-        
-        # Aggregate pattern intelligence across results
-        pattern_intelligence = aggregate_pattern_intelligence(all_results)
-        
-        # Update indexing status before returning results
-        await update_indexing_status()
-        
-        # Format results based on response_format and mode
-        timing_info['format_start'] = time.time()
-        
-        # Handle mode-specific responses
-        if mode == "quick":
-            # Quick mode: return just count and top result
-            total_count = len(all_results)  # Before we limited to 1
-            if response_format == "xml":
-                result_text = f"<quick_search>\n"
-                result_text += f"  <count>{total_count}</count>\n"
-                if all_results:
-                    top_result = all_results[0]
-                    result_text += f"  <top_result>\n"
-                    result_text += f"    <score>{top_result.score:.3f}</score>\n"
-                    result_text += f"    <excerpt>{escape(top_result.excerpt[:200])}</excerpt>\n"
-                    result_text += f"    <project>{escape(top_result.project_name)}</project>\n"
-                    result_text += f"    <conversation_id>{escape(top_result.conversation_id or '')}</conversation_id>\n"
-                    result_text += f"  </top_result>\n"
-                result_text += f"</quick_search>"
-                return result_text
-            else:
-                # Markdown format for quick mode
-                if all_results:
-                    return f"**Found {total_count} matches**\n\nTop result (score: {all_results[0].score:.3f}):\n{all_results[0].excerpt[:200]}"
-                else:
-                    return f"No matches found for '{query}'"
-        
-        elif mode == "summary":
-            # Summary mode: return aggregated insights without individual results
-            if not all_results:
-                return f"No conversations found to summarize for '{query}'"
-            
-            # Aggregate data
-            total_count = len(all_results)
-            avg_score = sum(r.score for r in all_results) / total_count
-            
-            # Extract common concepts and tools
-            all_concepts = []
-            all_tools = []
-            all_files = []
-            projects = set()
-            
-            for result in all_results:
-                if result.concepts:
-                    all_concepts.extend(result.concepts)
-                if result.tools_used:
-                    all_tools.extend(result.tools_used)
-                if result.files_analyzed:
-                    all_files.extend(result.files_analyzed)
-                projects.add(result.project_name)
-            
-            # Count frequencies
-            from collections import Counter
-            concept_counts = Counter(all_concepts).most_common(5)
-            tool_counts = Counter(all_tools).most_common(5)
-            
-            if response_format == "xml":
-                result_text = f"<search_summary>\n"
-                result_text += f"  <query>{escape(query)}</query>\n"
-                result_text += f"  <total_matches>{total_count}</total_matches>\n"
-                result_text += f"  <average_score>{avg_score:.3f}</average_score>\n"
-                result_text += f"  <projects_involved>{len(projects)}</projects_involved>\n"
-                if concept_counts:
-                    result_text += f"  <common_concepts>\n"
-                    for concept, count in concept_counts:
-                        result_text += f"    <concept count=\"{count}\">{escape(concept)}</concept>\n"
-                    result_text += f"  </common_concepts>\n"
-                if tool_counts:
-                    result_text += f"  <common_tools>\n"
-                    for tool, count in tool_counts:
-                        result_text += f"    <tool count=\"{count}\">{escape(tool)}</tool>\n"
-                    result_text += f"  </common_tools>\n"
-                result_text += f"</search_summary>"
-                return result_text
-            else:
-                # Markdown format for summary
-                result_text = f"## Summary for: {query}\n\n"
-                result_text += f"- **Total matches**: {total_count}\n"
-                result_text += f"- **Average relevance**: {avg_score:.3f}\n"
-                result_text += f"- **Projects involved**: {len(projects)}\n\n"
-                if concept_counts:
-                    result_text += "**Common concepts**:\n"
-                    for concept, count in concept_counts:
-                        result_text += f"- {concept} ({count} occurrences)\n"
-                if tool_counts:
-                    result_text += "\n**Common tools**:\n"
-                    for tool, count in tool_counts:
-                        result_text += f"- {tool} ({count} uses)\n"
-                return result_text
-        
-        # Continue with normal formatting for full mode
-        if response_format == "xml":
-            # Add upfront summary for immediate visibility (before collapsible XML)
-            upfront_summary = ""
-            
-            # Show result summary
-            if all_results:
-                score_info = "high" if all_results[0].score >= 0.85 else "good" if all_results[0].score >= 0.75 else "partial"
-                upfront_summary += f"🎯 RESULTS: {len(all_results)} matches ({score_info} relevance, top score: {all_results[0].score:.3f})\n"
-                
-                # Show performance with indexing status inline
-                total_time = time.time() - start_time
-                indexing_info = f" | 📊 {indexing_status['indexed_conversations']}/{indexing_status['total_conversations']} indexed" if indexing_status["percentage"] < 100.0 else ""
-                upfront_summary += f"⚡ PERFORMANCE: {int(total_time * 1000)}ms ({len(collections_to_search)} collections searched{indexing_info})\n"
-            else:
-                upfront_summary += f"❌ NO RESULTS: No conversations found matching '{query}'\n"
-            
-            # XML format (compact tags for performance)
-            result_text = upfront_summary + "\n<search>\n"
-            
-            # Add indexing status if not fully baselined - put key stats in opening tag for immediate visibility
-            if indexing_status["percentage"] < 95.0:
-                result_text += f'  <info status="indexing" progress="{indexing_status["percentage"]:.1f}%" backlog="{indexing_status["backlog_count"]}">\n'
-                result_text += f'    <message>📊 Indexing: {indexing_status["indexed_conversations"]}/{indexing_status["total_conversations"]} conversations ({indexing_status["percentage"]:.1f}% complete, {indexing_status["backlog_count"]} pending)</message>\n'
-                result_text += f"  </info>\n"
-            
-            # Add high-level result summary
-            if all_results:
-                # Count today's results
-                now = datetime.now(timezone.utc)
-                today_count = 0
-                yesterday_count = 0
-                week_count = 0
-                
-                for result in all_results:
-                    timestamp_clean = result.timestamp.replace('Z', '+00:00') if result.timestamp.endswith('Z') else result.timestamp
-                    timestamp_dt = datetime.fromisoformat(timestamp_clean)
-                    if timestamp_dt.tzinfo is None:
-                        timestamp_dt = timestamp_dt.replace(tzinfo=timezone.utc)
-                    
-                    days_ago = (now - timestamp_dt).days
-                    if days_ago == 0:
-                        today_count += 1
-                    elif days_ago == 1:
-                        yesterday_count += 1
-                    if days_ago <= 7:
-                        week_count += 1
-                
-                # Compact summary with key info in opening tag
-                time_info = ""
-                if today_count > 0:
-                    time_info = f"{today_count} today"
-                elif yesterday_count > 0:
-                    time_info = f"{yesterday_count} yesterday"
-                elif week_count > 0:
-                    time_info = f"{week_count} this week"
-                else:
-                    time_info = "older results"
-                
-                score_info = "high" if all_results[0].score >= 0.85 else "good" if all_results[0].score >= 0.75 else "partial"
-                
-                result_text += f'  <summary count="{len(all_results)}" relevance="{score_info}" recency="{time_info}" top-score="{all_results[0].score:.3f}">\n'
-                
-                # Short preview of top result
-                top_excerpt = all_results[0].excerpt[:100].strip()
-                if '...' not in top_excerpt:
-                    top_excerpt += "..."
-                result_text += f'    <preview>{top_excerpt}</preview>\n'
-                result_text += f"  </summary>\n"
-            else:
-                result_text += f"  <result-summary>\n"
-                result_text += f"    <headline>No matches found</headline>\n"
-                result_text += f"    <relevance>No conversations matched your query</relevance>\n"
-                result_text += f"  </result-summary>\n"
-            
-            result_text += f"  <meta>\n"
-            result_text += f"    <q>{query}</q>\n"
-            result_text += f"    <scope>{target_project if target_project != 'all' else 'all'}</scope>\n"
-            result_text += f"    <count>{len(all_results)}</count>\n"
-            if all_results:
-                result_text += f"    <range>{all_results[-1].score:.3f}-{all_results[0].score:.3f}</range>\n"
-            result_text += f"    <embed>{'local' if PREFER_LOCAL_EMBEDDINGS or not voyage_client else 'voyage'}</embed>\n"
-            
-            # Add timing metadata
-            total_time = time.time() - start_time
-            result_text += f"    <perf>\n"
-            result_text += f"      <ttl>{int(total_time * 1000)}</ttl>\n"
-            result_text += f"      <emb>{int((timing_info.get('embedding_end', 0) - timing_info.get('embedding_start', 0)) * 1000)}</emb>\n"
-            result_text += f"      <srch>{int((timing_info.get('search_all_end', 0) - timing_info.get('search_all_start', 0)) * 1000)}</srch>\n"
-            result_text += f"      <cols>{len(collections_to_search)}</cols>\n"
-            result_text += f"    </perf>\n"
-            result_text += f"  </meta>\n"
-            
-            result_text += "  <results>\n"
-            for i, result in enumerate(all_results):
-                result_text += f'    <r rank="{i+1}">\n'
-                result_text += f"      <s>{result.score:.3f}</s>\n"
-                result_text += f"      <p>{result.project_name}</p>\n"
-                
-                # Calculate relative time
-                timestamp_clean = result.timestamp.replace('Z', '+00:00') if result.timestamp.endswith('Z') else result.timestamp
-                timestamp_dt = datetime.fromisoformat(timestamp_clean)
-                # Ensure both datetimes are timezone-aware
-                if timestamp_dt.tzinfo is None:
-                    timestamp_dt = timestamp_dt.replace(tzinfo=timezone.utc)
-                now = datetime.now(timezone.utc)
-                days_ago = (now - timestamp_dt).days
-                if days_ago == 0:
-                    time_str = "today"
-                elif days_ago == 1:
-                    time_str = "yesterday"
-                else:
-                    time_str = f"{days_ago}d"
-                result_text += f"      <t>{time_str}</t>\n"
-                
-                if not brief:
-                    # Extract title from first line of excerpt
-                    excerpt_lines = result.excerpt.split('\n')
-                    title = excerpt_lines[0][:80] + "..." if len(excerpt_lines[0]) > 80 else excerpt_lines[0]
-                    result_text += f"      <title>{title}</title>\n"
-                    
-                    # Key finding - summarize the main point
-                    key_finding = result.excerpt[:100] + "..." if len(result.excerpt) > 100 else result.excerpt
-                    result_text += f"      <key-finding>{key_finding.strip()}</key-finding>\n"
-                
-                # Always include excerpt, but shorter in brief mode
-                if brief:
-                    brief_excerpt = result.excerpt[:100] + "..." if len(result.excerpt) > 100 else result.excerpt
-                    result_text += f"      <excerpt>{brief_excerpt.strip()}</excerpt>\n"
-                else:
-                    result_text += f"      <excerpt><![CDATA[{result.excerpt}]]></excerpt>\n"
-                
-                if result.conversation_id:
-                    result_text += f"      <cid>{result.conversation_id}</cid>\n"
-                
-                # Include raw data if requested
-                if include_raw and result.raw_payload:
-                    result_text += "      <raw>\n"
-                    result_text += f"        <txt><![CDATA[{result.raw_payload.get('text', '')}]]></txt>\n"
-                    result_text += f"        <id>{result.id}</id>\n"
-                    result_text += f"        <dist>{1 - result.score:.3f}</dist>\n"
-                    result_text += "        <meta>\n"
-                    for key, value in result.raw_payload.items():
-                        if key != 'text':
-                            result_text += f"          <{key}>{value}</{key}>\n"
-                    result_text += "        </meta>\n"
-                    result_text += "      </raw>\n"
-                
-                # Add patterns if they exist - with detailed logging
-                if result.code_patterns and isinstance(result.code_patterns, dict):
-                    logger.info(f"DEBUG: Point {result.id} has code_patterns dict with keys: {list(result.code_patterns.keys())}")
-                    patterns_to_show = []
-                    for category, patterns in result.code_patterns.items():
-                        if patterns and isinstance(patterns, list) and len(patterns) > 0:
-                            # Take up to 5 patterns from each category
-                            patterns_to_show.append((category, patterns[:5]))
-                            logger.info(f"DEBUG: Added category '{category}' with {len(patterns)} patterns")
-                    
-                    if patterns_to_show:
-                        logger.info(f"DEBUG: Adding patterns XML for point {result.id}")
-                        result_text += "      <patterns>\n"
-                        for category, patterns in patterns_to_show:
-                            # Escape both category name and pattern content for XML safety
-                            safe_patterns = ', '.join(escape(str(p)) for p in patterns)
-                            result_text += f"        <cat name=\"{escape(category)}\">{safe_patterns}</cat>\n"
-                        result_text += "      </patterns>\n"
-                    else:
-                        logger.info(f"DEBUG: Point {result.id} has code_patterns but no valid patterns to show")
-                else:
-                    logger.info(f"DEBUG: Point {result.id} has no patterns. code_patterns={result.code_patterns}, type={type(result.code_patterns)}")
-                
-                if result.files_analyzed and len(result.files_analyzed) > 0:
-                    result_text += f"      <files>{', '.join(result.files_analyzed[:5])}</files>\n"
-                if result.concepts and len(result.concepts) > 0:
-                    result_text += f"      <concepts>{', '.join(result.concepts[:5])}</concepts>\n"
-                
-                # Include structured metadata for agent consumption
-                # This provides clean, parsed fields that agents can easily use
-                if hasattr(result, 'raw_payload') and result.raw_payload:
-                    import json
-                    payload = result.raw_payload
-                    
-                    # Files section - structured for easy agent parsing
-                    files_analyzed = payload.get('files_analyzed', [])
-                    files_edited = payload.get('files_edited', [])
-                    if files_analyzed or files_edited:
-                        result_text += "      <files>\n"
-                        if files_analyzed:
-                            result_text += f"        <analyzed count=\"{len(files_analyzed)}\">"
-                            result_text += ", ".join(files_analyzed[:5])  # First 5 files
-                            if len(files_analyzed) > 5:
-                                result_text += f" ... and {len(files_analyzed)-5} more"
-                            result_text += "</analyzed>\n"
-                        if files_edited:
-                            result_text += f"        <edited count=\"{len(files_edited)}\">"
-                            result_text += ", ".join(files_edited[:5])  # First 5 files
-                            if len(files_edited) > 5:
-                                result_text += f" ... and {len(files_edited)-5} more"
-                            result_text += "</edited>\n"
-                        result_text += "      </files>\n"
-                    
-                    # Concepts section - clean list for agents
-                    concepts = payload.get('concepts', [])
-                    if concepts:
-                        result_text += f"      <concepts>{', '.join(concepts)}</concepts>\n"
-                    
-                    # Tools section - summarized with counts
-                    tools_used = payload.get('tools_used', [])
-                    if tools_used:
-                        # Count tool usage
-                        tool_counts = {}
-                        for tool in tools_used:
-                            tool_counts[tool] = tool_counts.get(tool, 0) + 1
-                        # Sort by frequency
-                        sorted_tools = sorted(tool_counts.items(), key=lambda x: x[1], reverse=True)
-                        tool_summary = ", ".join(f"{tool}({count})" for tool, count in sorted_tools[:5])
-                        if len(sorted_tools) > 5:
-                            tool_summary += f" ... and {len(sorted_tools)-5} more"
-                        result_text += f"      <tools>{tool_summary}</tools>\n"
-                    
-                    # Code patterns section - structured by category
-                    code_patterns = payload.get('code_patterns', {})
-                    if code_patterns:
-                        result_text += "      <code_patterns>\n"
-                        for category, patterns in code_patterns.items():
-                            if patterns:
-                                pattern_list = patterns if isinstance(patterns, list) else [patterns]
-                                # Clean up pattern names
-                                clean_patterns = []
-                                for p in pattern_list[:5]:
-                                    # Remove common prefixes like $FUNC, $VAR
-                                    clean_p = str(p).replace('$FUNC', '').replace('$VAR', '').strip()
-                                    if clean_p:
-                                        clean_patterns.append(clean_p)
-                                if clean_patterns:
-                                    result_text += f"        <{category}>{', '.join(clean_patterns)}</{category}>\n"
-                        result_text += "      </code_patterns>\n"
-                    
-                    # Pattern inheritance info - shows propagation details
-                    pattern_inheritance = payload.get('pattern_inheritance', {})
-                    if pattern_inheritance:
-                        source_chunk = pattern_inheritance.get('source_chunk', '')
-                        confidence = pattern_inheritance.get('confidence', 0)
-                        distance = pattern_inheritance.get('distance', 0)
-                        if source_chunk:
-                            result_text += f"      <pattern_source chunk=\"{source_chunk}\" confidence=\"{confidence:.2f}\" distance=\"{distance}\"/>\n"
-                    
-                    # Message stats for context
-                    msg_count = payload.get('message_count')
-                    total_length = payload.get('total_length')
-                    if msg_count or total_length:
-                        stats_attrs = []
-                        if msg_count:
-                            stats_attrs.append(f'messages="{msg_count}"')
-                        if total_length:
-                            stats_attrs.append(f'length="{total_length}"')
-                        result_text += f"      <stats {' '.join(stats_attrs)}/>\n"
-                    
-                    # Raw metadata dump for backwards compatibility
-                    # Kept minimal - only truly unique fields
-                    remaining_metadata = {}
-                    excluded_keys = {'text', 'conversation_id', 'timestamp', 'role', 'project', 'chunk_index',
-                                   'files_analyzed', 'files_edited', 'concepts', 'tools_used', 
-                                   'code_patterns', 'pattern_inheritance', 'message_count', 'total_length',
-                                   'chunking_version', 'chunk_method', 'chunk_overlap', 'migration_type'}
-                    for key, value in payload.items():
-                        if key not in excluded_keys and value is not None:
-                            if isinstance(value, set):
-                                value = list(value)
-                            remaining_metadata[key] = value
-                    
-                    if remaining_metadata:
-                        try:
-                            # Only include if there's actually extra data
-                            result_text += f"      <metadata_extra><![CDATA[{json.dumps(remaining_metadata, default=str)}]]></metadata_extra>\n"
-                        except:
-                            pass
-                
-                result_text += "    </r>\n"
-            result_text += "  </results>\n"
-            
-            # Add aggregated pattern intelligence section
-            if pattern_intelligence and pattern_intelligence.get('total_unique_patterns', 0) > 0:
-                result_text += "  <pattern_intelligence>\n"
-                
-                # Summary statistics
-                result_text += f"    <summary>\n"
-                result_text += f"      <unique_patterns>{pattern_intelligence['total_unique_patterns']}</unique_patterns>\n"
-                result_text += f"      <pattern_diversity>{pattern_intelligence['pattern_diversity_score']:.2f}</pattern_diversity>\n"
-                result_text += f"    </summary>\n"
-                
-                # Most common patterns
-                if pattern_intelligence.get('most_common_patterns'):
-                    result_text += "    <common_patterns>\n"
-                    for pattern, count in pattern_intelligence['most_common_patterns'][:5]:
-                        result_text += f"      <pattern count=\"{count}\">{pattern}</pattern>\n"
-                    result_text += "    </common_patterns>\n"
-                
-                # Pattern categories
-                if pattern_intelligence.get('category_coverage'):
-                    result_text += "    <categories>\n"
-                    for category, count in pattern_intelligence['category_coverage'].items():
-                        result_text += f"      <cat name=\"{category}\" count=\"{count}\"/>\n"
-                    result_text += "    </categories>\n"
-                
-                # Pattern combinations insight
-                if pattern_intelligence.get('pattern_combinations'):
-                    combos = pattern_intelligence['pattern_combinations']
-                    if combos.get('async_with_error_handling'):
-                        result_text += "    <insight>Async patterns combined with error handling detected</insight>\n"
-                    if combos.get('react_with_state'):
-                        result_text += "    <insight>React hooks with state management patterns detected</insight>\n"
-                
-                # Files referenced across results
-                if pattern_intelligence.get('files_referenced') and len(pattern_intelligence['files_referenced']) > 0:
-                    result_text += f"    <files_across_results>{', '.join(pattern_intelligence['files_referenced'][:10])}</files_across_results>\n"
-                
-                # Concepts discussed
-                if pattern_intelligence.get('concepts_discussed') and len(pattern_intelligence['concepts_discussed']) > 0:
-                    result_text += f"    <concepts_discussed>{', '.join(pattern_intelligence['concepts_discussed'][:10])}</concepts_discussed>\n"
-                
-                result_text += "  </pattern_intelligence>\n"
-            
-            result_text += "</search>"
-            
-        else:
-            # Markdown format (original)
-            result_text = f"Found {len(all_results)} relevant conversation(s) for '{query}':\n\n"
-            for i, result in enumerate(all_results):
-                result_text += f"**Result {i+1}** (Score: {result.score:.3f})\n"
-                # Handle timezone suffix 'Z' properly
-                timestamp_clean = result.timestamp.replace('Z', '+00:00') if result.timestamp.endswith('Z') else result.timestamp
-                result_text += f"Time: {datetime.fromisoformat(timestamp_clean).strftime('%Y-%m-%d %H:%M:%S')}\n"
-                result_text += f"Project: {result.project_name}\n"
-                result_text += f"Role: {result.role}\n"
-                result_text += f"Excerpt: {result.excerpt}\n"
-                result_text += "---\n\n"
-        
-        timing_info['format_end'] = time.time()
-        
-        # Log detailed timing breakdown
-        await ctx.debug(f"\n=== TIMING BREAKDOWN ===")
-        await ctx.debug(f"Total time: {(time.time() - start_time) * 1000:.1f}ms")
-        await ctx.debug(f"Embedding generation: {(timing_info.get('embedding_end', 0) - timing_info.get('embedding_start', 0)) * 1000:.1f}ms")
-        await ctx.debug(f"Get collections: {(timing_info.get('get_collections_end', 0) - timing_info.get('get_collections_start', 0)) * 1000:.1f}ms")
-        await ctx.debug(f"Search all collections: {(timing_info.get('search_all_end', 0) - timing_info.get('search_all_start', 0)) * 1000:.1f}ms")
-        await ctx.debug(f"Sorting results: {(timing_info.get('sort_end', 0) - timing_info.get('sort_start', 0)) * 1000:.1f}ms")
-        await ctx.debug(f"Formatting output: {(timing_info.get('format_end', 0) - timing_info.get('format_start', 0)) * 1000:.1f}ms")
-        
-        # Log per-collection timings
-        await ctx.debug(f"\n=== PER-COLLECTION TIMINGS ===")
-        for ct in collection_timings:
-            duration = (ct.get('end', 0) - ct.get('start', 0)) * 1000
-            status = "ERROR" if 'error' in ct else "OK"
-            await ctx.debug(f"{ct['name']}: {duration:.1f}ms ({status})")
-        
-        return result_text
-        
-    except Exception as e:
-        await ctx.error(f"Search failed: {str(e)}")
-        return f"Failed to search conversations: {str(e)}"
-
-
-@mcp.tool()
-async def store_reflection(
-    ctx: Context,
-    content: str = Field(description="The insight or reflection to store"),
-    tags: List[str] = Field(default=[], description="Tags to categorize this reflection")
-) -> str:
-    """Store an important insight or reflection for future reference."""
-    
-    try:
-        # Create reflections collection name
-        collection_name = f"reflections{get_collection_suffix()}"
-        
-        # Get current project context
-        cwd = os.environ.get('MCP_CLIENT_CWD', os.getcwd())
-        project_path = Path(cwd)
-        
-        # Extract project name from path
-        project_name = None
-        path_parts = project_path.parts
-        if 'projects' in path_parts:
-            idx = path_parts.index('projects')
-            if idx + 1 < len(path_parts):
-                # Get all parts after 'projects' to form the project name
-                # This handles cases like projects/Connectiva-App/connectiva-ai
-                project_parts = path_parts[idx + 1:]
-                project_name = '/'.join(project_parts)
-        
-        # If no project detected, use the last directory name
-        if not project_name:
-            project_name = project_path.name
-        
-        # Ensure collection exists
-        try:
-            collection_info = await qdrant_client.get_collection(collection_name)
-        except:
-            # Create collection if it doesn't exist
-            await qdrant_client.create_collection(
-                collection_name=collection_name,
-                vectors_config=VectorParams(
-                    size=get_embedding_dimension(),
-                    distance=Distance.COSINE
-                )
-            )
-            await ctx.debug(f"Created reflections collection: {collection_name}")
-        
-        # Generate embedding for the reflection
-        embedding = await generate_embedding(content)
-        
-        # Create point with metadata including project context
-        point_id = datetime.now().timestamp()
-        point = PointStruct(
-            id=int(point_id),
-            vector=embedding,
-            payload={
-                "text": content,
-                "tags": tags,
-                "timestamp": datetime.now().isoformat(),
-                "type": "reflection",
-                "role": "user_reflection",
-                "project": project_name,  # Add project context
-                "project_path": str(project_path)  # Add full path for reference
-            }
-        )
-        
-        # Store in Qdrant
-        await qdrant_client.upsert(
-            collection_name=collection_name,
-            points=[point]
-        )
-        
-        tags_str = ', '.join(tags) if tags else 'none'
-        return f"Reflection stored successfully with tags: {tags_str}"
-        
-    except Exception as e:
-        await ctx.error(f"Store failed: {str(e)}")
-        return f"Failed to store reflection: {str(e)}"
-
-
-@mcp.tool()
-async def quick_search(
-    ctx: Context,
-    query: str = Field(description="The search query to find semantically similar conversations"),
-    min_score: float = Field(default=0.3, description="Minimum similarity score (0-1)"),
-    project: Optional[str] = Field(default=None, description="Search specific project only. If not provided, searches current project based on working directory. Use 'all' to search across all projects.")
-) -> str:
-    """Quick search that returns only the count and top result for fast overview."""
-    # MCP architectural limitation: MCP tools cannot call other MCP tools
-    return """<error>
-MCP Architectural Limitation: This tool cannot directly call other MCP tools.
 
-To perform a quick search, please:
-1. Call reflect_on_past directly with limit=1 and brief=True
-2. Or use the reflection-specialist agent for quick searches
-
-This limitation exists because MCP tools can only be orchestrated by the client (Claude), 
-not by other tools within the MCP server.
-</error>"""
-
-
-@mcp.tool()
-async def search_summary(
-    ctx: Context,
-    query: str = Field(description="The search query to find semantically similar conversations"),
-    project: Optional[str] = Field(default=None, description="Search specific project only. If not provided, searches current project based on working directory. Use 'all' to search across all projects.")
-) -> str:
-    """Get aggregated insights from search results without individual result details."""
-    # MCP architectural limitation: MCP tools cannot call other MCP tools
-    # This is a fundamental constraint of the MCP protocol
-    return """<error>
-MCP Architectural Limitation: This tool cannot directly call other MCP tools.
-
-To get a search summary, please use the reflection-specialist agent instead:
-1. Call the reflection-specialist agent
-2. Ask it to provide a summary of search results for your query
-
-Alternative: Call reflect_on_past directly and analyze the results yourself.
-
-This limitation exists because MCP tools can only be orchestrated by the client (Claude), 
-not by other tools within the MCP server.
-</error>"""
-
-
-@mcp.tool()
-async def get_more_results(
-    ctx: Context,
-    query: str = Field(description="The original search query"),
-    offset: int = Field(default=3, description="Number of results to skip (for pagination)"),
-    limit: int = Field(default=3, description="Number of additional results to return"),
-    min_score: float = Field(default=0.3, description="Minimum similarity score (0-1)"),
-    project: Optional[str] = Field(default=None, description="Search specific project only")
-) -> str:
-    """Get additional search results after an initial search (pagination support)."""
-    # MCP architectural limitation: MCP tools cannot call other MCP tools
-    return """<error>
-MCP Architectural Limitation: This tool cannot directly call other MCP tools.
-
-To get more search results, please:
-1. Call reflect_on_past directly with a higher limit parameter
-2. Or use the reflection-specialist agent to handle pagination
-
-This limitation exists because MCP tools can only be orchestrated by the client (Claude), 
-not by other tools within the MCP server.
-</error>"""
-
-
-@mcp.tool()
-async def search_by_file(
-    ctx: Context,
-    file_path: str = Field(description="The file path to search for in conversations"),
-    limit: int = Field(default=10, description="Maximum number of results to return"),
-    project: Optional[str] = Field(default=None, description="Search specific project only. Use 'all' to search across all projects.")
-) -> str:
-    """Search for conversations that analyzed a specific file."""
-    global qdrant_client
-    
-    # Normalize file path
-    normalized_path = file_path.replace("\\", "/").replace("/Users/", "~/")
-    
-    # Determine which collections to search
-    # If no project specified, search all collections
-    collections = await get_all_collections() if not project else []
-    
-    if project and project != 'all':
-        # Filter collections for specific project - normalize first!
-        normalized_project = normalize_project_name(project)
-        project_hash = hashlib.md5(normalized_project.encode()).hexdigest()[:8]
-        collection_prefix = f"conv_{project_hash}_"
-        collections = [c for c in await get_all_collections() if c.startswith(collection_prefix)]
-    elif project == 'all':
-        collections = await get_all_collections()
-    
-    if not collections:
-        return "<search_by_file>\n<error>No collections found to search</error>\n</search_by_file>"
-    
-    # Prepare results
-    all_results = []
-    
-    for collection_name in collections:
-        try:
-            # Use scroll to get all points and filter manually
-            # Qdrant's array filtering can be tricky, so we'll filter in code
-            scroll_result = await qdrant_client.scroll(
-                collection_name=collection_name,
-                limit=1000,  # Get a batch
-                with_payload=True
-            )
-            
-            # Filter results that contain the file
-            for point in scroll_result[0]:
-                payload = point.payload
-                files_analyzed = payload.get('files_analyzed', [])
-                files_edited = payload.get('files_edited', [])
-                
-                # Check for exact match or if any file ends with the normalized path
-                file_match = False
-                for file in files_analyzed + files_edited:
-                    if file == normalized_path or file.endswith('/' + normalized_path) or file.endswith('\\' + normalized_path):
-                        file_match = True
-                        break
-                
-                if file_match:
-                    all_results.append({
-                        'score': 1.0,  # File match is always 1.0
-                        'payload': payload,
-                        'collection': collection_name
-                    })
-                
-        except Exception as e:
-            continue
-    
-    # Sort by timestamp (newest first)
-    all_results.sort(key=lambda x: x['payload'].get('timestamp', ''), reverse=True)
-    
-    # Format results
-    if not all_results:
-        return f"""<search_by_file>
-<query>{file_path}</query>
-<normalized_path>{normalized_path}</normalized_path>
-<message>No conversations found that analyzed this file</message>
-</search_by_file>"""
-    
-    results_text = []
-    for i, result in enumerate(all_results[:limit]):
-        payload = result['payload']
-        timestamp = payload.get('timestamp', 'Unknown')
-        conversation_id = payload.get('conversation_id', 'Unknown')
-        project = payload.get('project', 'Unknown')
-        text_preview = payload.get('text', '')[:200] + '...' if len(payload.get('text', '')) > 200 else payload.get('text', '')
-        
-        # Check if file was edited or just read
-        action = "edited" if normalized_path in payload.get('files_edited', []) else "analyzed"
-        
-        # Get related tools used
-        tool_summary = payload.get('tool_summary', {})
-        tools_used = ', '.join(f"{tool}({count})" for tool, count in tool_summary.items())
-        
-        results_text.append(f"""<result rank="{i+1}">
-<conversation_id>{conversation_id}</conversation_id>
-<project>{project}</project>
-<timestamp>{timestamp}</timestamp>
-<action>{action}</action>
-<tools_used>{tools_used}</tools_used>
-<preview>{text_preview}</preview>
-</result>""")
-    
-    return f"""<search_by_file>
-<query>{file_path}</query>
-<normalized_path>{normalized_path}</normalized_path>
-<count>{len(all_results)}</count>
-<results>
-{''.join(results_text)}
-</results>
-</search_by_file>"""
-
-
-@mcp.tool()
-async def search_by_concept(
-    ctx: Context,
-    concept: str = Field(description="The concept to search for (e.g., 'security', 'docker', 'testing')"),
-    include_files: bool = Field(default=True, description="Include file information in results"),
-    limit: int = Field(default=10, description="Maximum number of results to return"),
-    project: Optional[str] = Field(default=None, description="Search specific project only. Use 'all' to search across all projects.")
-) -> str:
-    """Search for conversations about a specific development concept."""
-    global qdrant_client
-    
-    # Generate embedding for the concept
-    embedding = await generate_embedding(concept)
-    
-    # Determine which collections to search
-    # If no project specified, search all collections
-    collections = await get_all_collections() if not project else []
-    
-    if project and project != 'all':
-        # Filter collections for specific project
-        normalized_project = normalize_project_name(project)
-        project_hash = hashlib.md5(normalized_project.encode()).hexdigest()[:8]
-        collection_prefix = f"conv_{project_hash}_"
-        collections = [c for c in await get_all_collections() if c.startswith(collection_prefix)]
-    elif project == 'all':
-        collections = await get_all_collections()
-    
-    if not collections:
-        return "<search_by_concept>\n<error>No collections found to search</error>\n</search_by_concept>"
-    
-    # First, check metadata health
-    metadata_found = False
-    total_points_checked = 0
-    
-    for collection_name in collections[:3]:  # Sample first 3 collections
-        try:
-            sample_points, _ = await qdrant_client.scroll(
-                collection_name=collection_name,
-                limit=10,
-                with_payload=True
-            )
-            total_points_checked += len(sample_points)
-            for point in sample_points:
-                if 'concepts' in point.payload and point.payload['concepts']:
-                    metadata_found = True
-                    break
-            if metadata_found:
-                break
-        except:
-            continue
-    
-    # Search all collections
-    all_results = []
-    
-    # If metadata exists, try metadata-based search first
-    if metadata_found:
-        for collection_name in collections:
-            try:
-                # Hybrid search: semantic + concept filter
-                results = await qdrant_client.search(
-                    collection_name=collection_name,
-                    query_vector=embedding,
-                    query_filter=models.Filter(
-                        should=[
-                            models.FieldCondition(
-                                key="concepts",
-                                match=models.MatchAny(any=[concept.lower()])
-                            )
-                        ]
-                    ),
-                    limit=limit * 2,  # Get more results for better filtering
-                    with_payload=True
-                )
-                
-                for point in results:
-                    payload = point.payload
-                    # Boost score if concept is in the concepts list
-                    score_boost = 0.2 if concept.lower() in payload.get('concepts', []) else 0.0
-                    all_results.append({
-                        'score': float(point.score) + score_boost,
-                        'payload': payload,
-                        'collection': collection_name,
-                        'search_type': 'metadata'
-                    })
-                    
-            except Exception as e:
-                # Log unexpected errors but continue with other collections
-                logger.debug(f"Error searching collection {collection_name}: {e}")
-                continue
-    
-    # If no results from metadata search OR no metadata exists, fall back to semantic search
-    if not all_results:
-        await ctx.debug(f"Falling back to semantic search for concept: {concept}")
-        
-        for collection_name in collections:
-            try:
-                # Pure semantic search without filters
-                results = await qdrant_client.search(
-                    collection_name=collection_name,
-                    query_vector=embedding,
-                    limit=limit,
-                    score_threshold=0.5,  # Lower threshold for broader results
-                    with_payload=True
-                )
-                
-                for point in results:
-                    all_results.append({
-                        'score': float(point.score),
-                        'payload': point.payload,
-                        'collection': collection_name,
-                        'search_type': 'semantic'
-                    })
-                    
-            except Exception as e:
-                # Log unexpected errors but continue with other collections
-                logger.debug(f"Error searching collection {collection_name}: {e}")
-                continue
-    
-    # Sort by score and limit
-    all_results.sort(key=lambda x: x['score'], reverse=True)
-    all_results = all_results[:limit]
-    
-    # Format results
-    if not all_results:
-        metadata_status = "with metadata" if metadata_found else "NO METADATA FOUND"
-        return f"""<search_by_concept>
-<concept>{concept}</concept>
-<metadata_health>{metadata_status} (checked {total_points_checked} points)</metadata_health>
-<message>No conversations found about this concept. {'Try running: python scripts/delta-metadata-update.py' if not metadata_found else 'Try different search terms.'}</message>
-</search_by_concept>"""
-    
-    results_text = []
-    for i, result in enumerate(all_results):
-        payload = result['payload']
-        score = result['score']
-        timestamp = payload.get('timestamp', 'Unknown')
-        conversation_id = payload.get('conversation_id', 'Unknown')
-        project = payload.get('project', 'Unknown')
-        concepts = payload.get('concepts', [])
-        
-        # Get text preview
-        text_preview = payload.get('text', '')[:200] + '...' if len(payload.get('text', '')) > 200 else payload.get('text', '')
-        
-        # File information
-        files_info = ""
-        if include_files:
-            files_analyzed = payload.get('files_analyzed', [])[:5]
-            if files_analyzed:
-                files_info = f"\n<files_analyzed>{', '.join(files_analyzed)}</files_analyzed>"
-        
-        # Related concepts
-        related_concepts = [c for c in concepts if c != concept.lower()][:5]
-        
-        results_text.append(f"""<result rank="{i+1}">
-<score>{score:.3f}</score>
-<conversation_id>{conversation_id}</conversation_id>
-<project>{project}</project>
-<timestamp>{timestamp}</timestamp>
-<concepts>{', '.join(concepts)}</concepts>
-<related_concepts>{', '.join(related_concepts)}</related_concepts>{files_info}
-<preview>{text_preview}</preview>
-</result>""")
-    
-    # Determine if this was a fallback search
-    used_fallback = any(r.get('search_type') == 'semantic' for r in all_results)
-    metadata_status = "with metadata" if metadata_found else "NO METADATA FOUND"
-    
-    return f"""<search_by_concept>
-<concept>{concept}</concept>
-<metadata_health>{metadata_status} (checked {total_points_checked} points)</metadata_health>
-<search_type>{'fallback_semantic' if used_fallback else 'metadata_based'}</search_type>
-<count>{len(all_results)}</count>
-<results>
-{''.join(results_text)}
-</results>
-</search_by_concept>"""
-
-
-# Debug output
-print(f"[DEBUG] FastMCP server created with name: {mcp.name}")
-
-@mcp.tool()
-async def get_full_conversation(
-    ctx: Context,
-    conversation_id: str = Field(description="The conversation ID from search results (cid)"),
-    project: Optional[str] = Field(default=None, description="Optional project name to help locate the file")
-) -> str:
-    """Get the full JSONL conversation file path for a conversation ID.
-    This allows agents to read complete conversations instead of truncated excerpts."""
-    
-    # Base path for Claude conversations
-    base_path = Path.home() / '.claude/projects'
-    
-    # Build list of directories to search
-    search_dirs = []
-    
-    if project:
-        # Try various project directory name formats
-        sanitized_project = project.replace('/', '-')
-        search_dirs.extend([
-            base_path / project,
-            base_path / sanitized_project,
-            base_path / f"-Users-*-projects-{project}",
-            base_path / f"-Users-*-projects-{sanitized_project}"
-        ])
-    else:
-        # Search all project directories
-        search_dirs = list(base_path.glob("*"))
-    
-    # Search for the JSONL file
-    jsonl_path = None
-    for search_dir in search_dirs:
-        if not search_dir.is_dir():
-            continue
-            
-        potential_path = search_dir / f"{conversation_id}.jsonl"
-        if potential_path.exists():
-            jsonl_path = potential_path
-            break
-    
-    if not jsonl_path:
-        # Try searching all directories as fallback
-        for proj_dir in base_path.glob("*"):
-            if proj_dir.is_dir():
-                potential_path = proj_dir / f"{conversation_id}.jsonl"
-                if potential_path.exists():
-                    jsonl_path = potential_path
-                    break
-    
-    if not jsonl_path:
-        return f"""<full_conversation>
-<conversation_id>{conversation_id}</conversation_id>
-<status>not_found</status>
-<message>Conversation file not found. Searched {len(search_dirs)} directories.</message>
-<hint>Try using the project parameter or check if the conversation ID is correct.</hint>
-</full_conversation>"""
-    
-    # Get file stats
-    file_stats = jsonl_path.stat()
-    
-    # Count messages
-    try:
-        with open(jsonl_path, 'r', encoding='utf-8') as f:
-            message_count = sum(1 for _ in f)
-    except:
-        message_count = 0
-    
-    return f"""<full_conversation>
-<conversation_id>{conversation_id}</conversation_id>
-<status>found</status>
-<file_path>{jsonl_path}</file_path>
-<file_size>{file_stats.st_size}</file_size>
-<message_count>{message_count}</message_count>
-<project>{jsonl_path.parent.name}</project>
-<instructions>
-You can now use the Read tool to read the full conversation from:
-{jsonl_path}
-
-Each line in the JSONL file is a separate message with complete content.
-This gives you access to:
-- Complete code blocks (not truncated)
-- Full problem descriptions and solutions
-- Entire debugging sessions
-- Complete architectural decisions and discussions
-</instructions>
-</full_conversation>"""
-
-
-@mcp.tool()
-async def get_next_results(
-    ctx: Context,
-    query: str = Field(description="The original search query"),
-    offset: int = Field(default=3, description="Number of results to skip (for pagination)"),
-    limit: int = Field(default=3, description="Number of additional results to return"),
-    min_score: float = Field(default=0.3, description="Minimum similarity score (0-1)"),
-    project: Optional[str] = Field(default=None, description="Search specific project only")
-) -> str:
-    """Get additional search results after an initial search (pagination support)."""
-    global qdrant_client, embedding_manager
-    
-    try:
-        # Generate embedding for the query
-        embedding = await generate_embedding(query)
-        
-        # Determine which collections to search
-        if project == "all" or not project:
-            # Search all collections if project is "all" or not specified
-            collections = await get_all_collections()
-        else:
-            # Search specific project - normalize first!
-            all_collections = await get_all_collections()
-            normalized_project = normalize_project_name(project)
-            project_hash = hashlib.md5(normalized_project.encode()).hexdigest()[:8]
-            collections = [
-                c for c in all_collections 
-                if c.startswith(f"conv_{project_hash}_")
-            ]
-            if not collections:
-                # Fall back to searching all collections
-                collections = all_collections
-        
-        if not collections:
-            return """<next_results>
-<error>No collections available to search</error>
-</next_results>"""
-        
-        # Collect all results from all collections
-        all_results = []
-        for collection_name in collections:
-            try:
-                # Check if collection exists
-                collection_info = await qdrant_client.get_collection(collection_name)
-                if not collection_info:
-                    continue
-                
-                # Search with reasonable limit to account for offset
-                max_search_limit = 100  # Define a reasonable cap
-                search_limit = min(offset + limit + 10, max_search_limit)
-                results = await qdrant_client.search(
-                    collection_name=collection_name,
-                    query_vector=embedding,
-                    limit=search_limit,
-                    score_threshold=min_score
-                )
-                
-                for point in results:
-                    payload = point.payload
-                    score = float(point.score)
-                    
-                    # Apply time-based decay if enabled
-                    use_decay_bool = ENABLE_MEMORY_DECAY  # Use global default
-                    if use_decay_bool and 'timestamp' in payload:
-                        try:
-                            timestamp = datetime.fromisoformat(payload['timestamp'].replace('Z', '+00:00'))
-                            age_days = (datetime.now(timezone.utc) - timestamp).total_seconds() / (24 * 60 * 60)
-                            # Use consistent half-life formula: decay = exp(-ln(2) * age / half_life)
-                            ln2 = math.log(2)
-                            decay_factor = math.exp(-ln2 * age_days / DECAY_SCALE_DAYS)
-                            # Apply multiplicative formula: score * ((1 - weight) + weight * decay)
-                            score = score * ((1 - DECAY_WEIGHT) + DECAY_WEIGHT * decay_factor)
-                        except (ValueError, TypeError) as e:
-                            # Log but continue - timestamp format issue shouldn't break search
-                            logger.debug(f"Failed to apply decay for timestamp {payload.get('timestamp')}: {e}")
-                    
-                    all_results.append({
-                        'score': score,
-                        'payload': payload,
-                        'collection': collection_name
-                    })
-                    
-            except Exception as e:
-                # Log unexpected errors but continue with other collections
-                logger.debug(f"Error searching collection {collection_name}: {e}")
-                continue
-        
-        # Sort by score
-        all_results.sort(key=lambda x: x['score'], reverse=True)
-        
-        # Apply pagination
-        paginated_results = all_results[offset:offset + limit]
-        
-        if not paginated_results:
-            return f"""<next_results>
-<query>{query}</query>
-<offset>{offset}</offset>
-<status>no_more_results</status>
-<message>No additional results found beyond offset {offset}</message>
-</next_results>"""
-        
-        # Format results
-        results_text = []
-        for i, result in enumerate(paginated_results, start=offset + 1):
-            payload = result['payload']
-            score = result['score']
-            timestamp = payload.get('timestamp', 'Unknown')
-            conversation_id = payload.get('conversation_id', 'Unknown')
-            project = payload.get('project', 'Unknown')
-            
-            # Get text preview (store text once to avoid multiple calls)
-            text = payload.get('text', '')
-            text_preview = text[:300] + '...' if len(text) > 300 else text
-            
-            results_text.append(f"""
-<result index="{i}">
-  <score>{score:.3f}</score>
-  <timestamp>{timestamp}</timestamp>
-  <project>{project}</project>
-  <conversation_id>{conversation_id}</conversation_id>
-  <preview>{text_preview}</preview>
-</result>""")
-        
-        # Check if there are more results available
-        has_more = len(all_results) > (offset + limit)
-        next_offset = offset + limit if has_more else None
-        
-        return f"""<next_results>
-<query>{query}</query>
-<offset>{offset}</offset>
-<limit>{limit}</limit>
-<count>{len(paginated_results)}</count>
-<total_available>{len(all_results)}</total_available>
-<has_more>{has_more}</has_more>
-{f'<next_offset>{next_offset}</next_offset>' if next_offset else ''}
-<results>{''.join(results_text)}
-</results>
-</next_results>"""
-        
-    except Exception as e:
-        await ctx.error(f"Pagination failed: {str(e)}")
-        return f"""<next_results>
-<error>Failed to get next results: {str(e)}</error>
-</next_results>"""
+# Register temporal tools after all functions are defined
+register_temporal_tools(
+    mcp,
+    qdrant_client,
+    QDRANT_URL,
+    get_all_collections,
+    generate_embedding,
+    initialize_embeddings,
+    normalize_project_name
+)
+print(f"[INFO] Temporal tools registered", file=sys.stderr)
+
+# Register search tools
+def get_embedding_manager():
+    """Factory function to get the current embedding manager."""
+    from .embedding_manager import get_embedding_manager as get_em
+    return get_em()
+
+# Initialize ProjectResolver for collection name mapping
+# ProjectResolver needs a sync client, not async
+from qdrant_client import QdrantClient as SyncQdrantClient
+sync_qdrant_client = SyncQdrantClient(url=QDRANT_URL)
+project_resolver = ProjectResolver(sync_qdrant_client)
+
+register_search_tools(
+    mcp,
+    qdrant_client,
+    QDRANT_URL,
+    get_embedding_manager,
+    normalize_project_name,
+    ENABLE_MEMORY_DECAY,
+    DECAY_WEIGHT,
+    DECAY_SCALE_DAYS,
+    USE_NATIVE_DECAY,
+    NATIVE_DECAY_AVAILABLE,
+    decay_manager,
+    project_resolver  # Pass the resolver
+)
 
+# Register reflection tools
+register_reflection_tools(
+    mcp,
+    qdrant_client,
+    QDRANT_URL,
+    get_embedding_manager,
+    normalize_project_name
+)
 
 # Run the server
 if __name__ == "__main__":