MCP Memory SQLite

# Feature Roadmap This document tracks potential future enhancements for mcp-memory-sqlite based on user feedback and LLM evaluation. **Current Status:** v0.0.4 - Text search with relevance ranking (8.5/10 rating from LLM evaluation) ## Recently Completed ✅ ### v0.0.4 - Vector Search Removal & Text Search Optimization - ✅ Removed non-functional vector search code - ✅ Added relevance-based search ranking (name > type > observation) - ✅ Added result limiting (default 10, max 50) - ✅ Added `get_entity_with_relations` for graph exploration - ✅ Optimized for LLM context efficiency (~12KB saved per entity) - ✅ Comprehensive documentation update **Impact:** Tool rated 8.5/10 by LLM evaluation, with 5/5 search quality rating. ## High Priority 🔥 ### Bulk Operations **Status:** Proposed **Priority:** High **Effort:** Small Add batch entity/relationship creation for efficiency: ```typescript // Current: Must create entities one array at a time create_entities({ entities: [...] }) create_relations({ relations: [...] }) // Proposed: Single transaction for large imports bulk_import({ entities: [...], // Hundreds of entities relations: [...], // Hundreds of relations }) ``` **Use case:** Importing existing knowledge bases, initializing project context **Benefits:** - Faster imports (single transaction) - Atomic operations (all or nothing) - Better for large-scale setup --- ### Relationship Metadata **Status:** Proposed **Priority:** Medium **Effort:** Small Add optional descriptions/properties to relations: ```typescript interface Relation { from: string; to: string; relationType: string; metadata?: { description?: string; strength?: number; // 0.0-1.0 confidence/importance created_at?: string; tags?: string[]; }; } ``` **Use case:** Richer context for relationships, temporal tracking **Benefits:** - More informative graph exploration - Track when/why relationships were created - Priority/importance for decision-making --- ## Medium Priority 📊 ### Complex Query Support **Status:** Proposed **Priority:** Medium **Effort:** Large Add query language for complex searches: ```typescript // Examples: search_advanced({ filters: { entityType: "person", hasRelation: { type: "works_on", target: "TaskFlow" }, observations: { contains: "frontend" } }, limit: 10 }) // Or find paths between entities: find_path({ from: "Alice", to: "PostgreSQL", maxDepth: 3 }) ``` **Use case:** Advanced knowledge graph queries, relationship discovery **Benefits:** - More powerful searches - Relationship inference - Complex decision support **Challenges:** - Query language design - Performance optimization - Increased API complexity --- ### Entity Versioning **Status:** Proposed **Priority:** Medium **Effort:** Large Track entity changes over time: ```typescript interface EntityVersion { entity: Entity; version: number; modified_at: string; changes: { added_observations?: string[]; removed_observations?: string[]; }; } // API: get_entity_history(name: string): EntityVersion[] restore_entity_version(name: string, version: number) ``` **Use case:** Audit trail, undo changes, temporal analysis **Benefits:** - Change tracking - Rollback capability - Understanding evolution **Challenges:** - Storage overhead - Migration complexity - UI/API design --- ## Low Priority 💡 ### Entity Merging **Status:** Proposed **Priority:** Low **Effort:** Medium Handle duplicate/similar entities: ```typescript merge_entities({ primary: "PostgreSQL", duplicates: ["postgres", "PostgresDB"], strategy: "keep_all_observations" // or "primary_only" }) ``` **Use case:** Deduplication, data cleanup **Benefits:** - Cleaner knowledge graph - Better search results - Easier maintenance --- ### Namespace/Project Separation **Status:** Proposed **Priority:** Low **Effort:** Large Support multiple isolated knowledge graphs: ```typescript // Current: Single global graph create_entities({ entities: [...] }) // Proposed: Project-scoped graphs create_entities({ namespace: "project_taskflow", entities: [...] }) search_nodes({ namespace: "project_taskflow", // or "all" query: "frontend" }) ``` **Use case:** Multiple projects, context isolation **Benefits:** - Prevents cross-project pollution - Clearer boundaries - Better organization **Challenges:** - Significant schema changes - Migration complexity - API design impact --- ### Export/Import **Status:** Proposed **Priority:** Low **Effort:** Small Export and import knowledge graphs: ```typescript export_graph({ format: "json", // or "graphml", "cypher" namespace?: "project_taskflow" }): string import_graph({ format: "json", data: string, merge_strategy: "replace" | "merge" }) ``` **Use case:** Backup, sharing, migration **Benefits:** - Portability - Backup/restore - Sharing knowledge bases --- ## Won't Do ❌ ### Vector/Semantic Search **Status:** Rejected **Reason:** Non-functional in practice, wastes context window Vector search was removed in v0.0.4 because: - LLMs cannot generate embeddings - 12KB context waste per entity - Text search with relevance ranking achieves similar results - Adds unnecessary complexity **Alternative:** Continue improving text search (fuzzy matching, relevance ranking, synonyms) --- ### AI-Powered Features **Status:** Not Planned **Reason:** Out of scope Features like auto-relationship detection, entity extraction from text, or auto-summarization are: - Better handled by the LLM using the tool - Outside the scope of a memory tool - Would require external API dependencies **Alternative:** Let the LLM handle intelligence, tool handles storage --- ## Contributing These features are suggestions based on user feedback. If you'd like to implement any: 1. **Open an issue** to discuss the approach 2. **Check for acceptance** before starting large changes 3. **Submit a PR** with tests and documentation **Priority Guidelines:** - High priority: Core functionality improvements, clear user value - Medium priority: Nice-to-have features with good use cases - Low priority: Edge cases or complex features with limited benefit --- ## Evaluation Criteria New features should meet these standards: ✅ **User value:** Solves a real LLM memory problem ✅ **Context efficient:** Doesn't bloat LLM context window ✅ **Simple API:** Easy for LLMs to use correctly ✅ **Well tested:** Includes tests and documentation ✅ **Maintainable:** Doesn't add significant complexity --- **Last updated:** Based on v0.0.4 LLM evaluation (8.5/10 rating) **Feedback welcome:** [Open an issue](https://github.com/spences10/mcp-memory-sqlite/issues) with feature suggestions or use case descriptions.