MCP Memory Service

# Timestamp Recall Fix Summary ## Issue Description The memory recall functionality in mcp-memory-service was experiencing issues with timestamp-based queries. Memories stored with precise timestamps (including sub-second precision) were being retrieved incorrectly or not at all when using time-based recall functions. ## Root Cause The issue was caused by timestamps being converted to integers at multiple points in the codebase: 1. **Storage**: In `ChromaMemoryStorage._optimize_metadata_for_chroma()`, timestamps were being stored as `int(memory.created_at)` 2. **Querying**: In the `recall()` method, timestamp comparisons were using `int(start_timestamp)` and `int(end_timestamp)` 3. **Memory Model**: In `Memory.to_dict()`, the timestamp field was being converted to `int(self.created_at)` This integer conversion caused loss of sub-second precision, making all memories within the same second indistinguishable by timestamp. ## Changes Made ### 1. Fixed Timestamp Storage (chroma.py) Changed line 949 in `_optimize_metadata_for_chroma()`: ```python # Before: "timestamp": int(memory.created_at), # After: "timestamp": float(memory.created_at), # Changed from int() to float() ``` ### 2. Fixed Timestamp Queries (chroma.py) Changed lines 739 and 743 in the `recall()` method: ```python # Before: where_clause["$and"].append({"timestamp": {"$gte": int(start_timestamp)}}) where_clause["$and"].append({"timestamp": {"$lte": int(end_timestamp)}}) # After: where_clause["$and"].append({"timestamp": {"$gte": float(start_timestamp)}}) where_clause["$and"].append({"timestamp": {"$lte": float(end_timestamp)}}) ``` ### 3. Fixed Memory Model (memory.py) Changed line 161 in `Memory.to_dict()`: ```python # Before: "timestamp": int(self.created_at), # Legacy timestamp (int) # After: "timestamp": float(self.created_at), # Changed from int() to preserve precision ``` ### 4. Fixed Date Parsing Order (time_parser.py) Moved the full ISO date pattern check before the specific date pattern check to prevent "2024-06-15" from being incorrectly parsed as "24-06-15". ## Tests Added ### 1. Timestamp Recall Tests (`tests/test_timestamp_recall.py`) - Tests for timestamp precision storage - Tests for natural language time parsing - Tests for various recall scenarios (yesterday, last week, specific dates) - Tests for combined semantic and time-based recall - Edge case tests ### 2. Time Parser Tests (`tests/test_time_parser.py`) - Comprehensive tests for all supported time expressions - Tests for relative dates (yesterday, 3 days ago, last week) - Tests for specific date formats (MM/DD/YYYY, YYYY-MM-DD) - Tests for seasons, holidays, and named periods - Tests for time extraction from queries ## Verification The fix has been verified with: 1. Unit tests covering individual components 2. Integration tests demonstrating end-to-end functionality 3. Precision tests showing that sub-second timestamps are now preserved ## Impact - Memories can now be recalled with precise timestamp filtering - Sub-second precision is maintained throughout storage and retrieval - Natural language time expressions work correctly - No breaking changes to existing functionality ## Recommendations 1. Consider adding database migration for existing memories with integer timestamps 2. Monitor performance impact of float vs integer comparisons in large datasets 3. Add documentation about supported time expressions for users