Skip to content

CHANGELOG.md

Latest commit

 

History

History
444 lines (359 loc) · 16.7 KB

CHANGELOG.md

File metadata and controls

444 lines (359 loc) · 16.7 KB

Changelog

All notable changes to Recall (formerly MCP Memory Server) are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[1.6.0] - 2025-10-04

Added

  • Time Window Context Retrieval - Get all memories from specific time periods

    • get_time_window_context tool for time-based memory retrieval
    • Multiple time specifications: hours, minutes, or explicit timestamp ranges
    • Three output formats: Markdown (default), JSON, and plain text
    • Four grouping options: chronological, by type, by importance, or by tags
    • Filtering by minimum importance and context types
    • Perfect for building context files from work sessions
    • Respects workspace modes (isolated/global/hybrid)

  • Best Practices Documentation - Context bloat prevention

    • Added guidelines for selective memory storage
    • Clear examples of what to store vs. avoid
    • Added to both README.md and CLAUDE.md
    • Helps users avoid indiscriminate memory storage

Technical

  • New GetTimeWindowContextSchema in types.ts
  • New getMemoriesByTimeWindow() method in MemoryStore
  • Helper functions for formatting output (JSON, Markdown, Text)
  • Uses Redis ZRANGEBYSCORE for efficient time-range queries

Testing

  • New test-time-window.js test suite with 8 tests
  • All tests passing
  • Test coverage for time-based retrieval, filtering, grouping

Tool Count

  • 28 tools (was 27 in v1.5.0)

[1.5.0] - 2025-10-03

Added

  • Memory Versioning & History - Track changes over time

    • Automatic versioning on memory updates (up to 50 versions per memory)
    • Version tracking includes content, context_type, importance, tags, timestamp
    • System-generated and user-generated versions with change reasons
    • MemoryVersion schema with created_by field ('user' or 'system')
    • Version history stored in Redis sorted sets by timestamp

  • Version Tools (2 new tools)

    • get_memory_history - Retrieve version history for a memory
    • rollback_memory - Rollback memory to a previous version (preserves relationships)

  • Memory Templates - Reusable memory patterns

    • Template system with variable placeholders ({{variable}} syntax)
    • Built-in and workspace-specific templates
    • Support for default tags, importance, and context_type
    • Template variable validation and replacement

  • Template Tools (3 new tools)

    • create_template - Create new memory template with placeholders
    • create_from_template - Instantiate memory from template
    • list_templates - List all available templates (workspace + builtin)

  • Advanced Search - Enhanced search capabilities

    • Fuzzy search with word-matching boost (up to 20% similarity increase)
    • Regex pattern matching for content filtering
    • Category filtering for organized search

  • Memory Categories - Organize memories by category

    • Category field added to memory schema (optional)
    • Category indexes for fast filtering
    • Automatic category tracking with last-used timestamps

  • Category Tools (3 new tools)

    • set_memory_category - Assign category to a memory
    • list_categories - List all categories with memory counts
    • get_memories_by_category - Retrieve all memories in a category

Enhanced

  • search_memories now supports:
    • category parameter for filtering by category
    • fuzzy boolean for fuzzy search
    • regex parameter for pattern matching
  • createMemory and updateMemory support category field
  • All memory operations maintain backward compatibility

Technical

  • Added version keys to Redis: memory:{id}:versions, memory:{id}:version:{versionId}
  • Added template keys: template:{id}, templates:all, builtin:templates:all
  • Added category keys: memory:{id}:category, category:{name}, categories:all
  • Version limit enforced at 50 per memory (ZREMRANGEBYRANK for cleanup)
  • Template variable replacement with {{placeholder}} syntax
  • Fuzzy search uses word-level matching with configurable boost
  • Category indexes support both workspace and global scopes

Use Cases

  • Track evolution of code patterns and decisions over time
  • Rollback incorrect updates while preserving relationships
  • Create standardized memory templates for team workflows
  • Organize memories by project, feature, or domain with categories
  • Use advanced search filters to find specific patterns

Tools: 27 total (6 core + 3 smart + 4 advanced + 2 global + 4 relationships + 2 versions + 3 templates + 3 categories) Resources: 17 total (unchanged from v1.4.0) Prompts: 1 (workspace_context)

[1.4.0] - 2025-10-03

Added

  • Memory Relationships - Link related memories to create knowledge graphs

    • RelationshipType enum with 7 relationship types:
      • relates_to - Generic connection
      • parent_of / child_of - Hierarchical relationships
      • references - One memory references another
      • supersedes - One memory replaces another
      • implements - Implementation of a pattern/concept
      • example_of - Example demonstrating a pattern
    • Bidirectional relationship tracking (outgoing/incoming/both)
    • Graph traversal with configurable depth (1-5 levels)
    • Circular reference protection
    • Global relationship support (inherits from memory scope)

  • Relationship Tools (4 new tools)

    • link_memories - Create relationships between memories
    • get_related_memories - Traverse relationship graph with depth control
    • unlink_memories - Remove relationships
    • get_memory_graph - Get full graph structure with max nodes limit

  • Relationship Resources (3 new resources)

    • memory://relationships?limit=100 - List all relationships
    • memory://memory/{id}/related?depth=1&direction=both - Get related memories
    • memory://graph/{id}?depth=2&max_nodes=50 - Get memory graph structure

Technical

  • Added MemoryRelationship interface with metadata support
  • Added relationship Redis keys for workspace and global scopes
  • Implemented createRelationship(), getRelationship(), deleteRelationship() in MemoryStore
  • Implemented getRelatedMemories() with graph traversal algorithm
  • Implemented getMemoryGraph() for full graph construction
  • Added serializeRelationship() and deserializeRelationship() helpers
  • Relationship scope determined by memory scope (global + global = global)

Use Cases

  • Link code patterns to their implementations
  • Connect decisions to their rationale and consequences
  • Build knowledge graphs of related concepts
  • Track feature evolution across related memories
  • Create hierarchical memory structures

Tools: 19 total (6 core + 3 smart + 4 advanced + 2 global + 4 relationships) Resources: 17 total (9 workspace + 5 global + 3 relationships) Prompts: 1 (workspace_context)

[1.3.0] - 2025-10-03

Added

  • Global Memories - Cross-workspace memory sharing
    • is_global field on memories - Mark memories as accessible across all workspaces
    • workspace_id field - Track workspace origin for each memory
    • Three workspace modes via WORKSPACE_MODE environment variable:
      • isolated (default) - Workspace-only, no cross-workspace access
      • global - All memories shared globally, no workspace isolation
      • hybrid - Both workspace AND global memories with smart weighting
    • Hybrid search weighting: workspace memories 1.0x, global memories 0.9x (prefer local context)
  • Global Memory Tools (2 new tools)
    • convert_to_global - Convert workspace-specific memory to global
    • convert_to_workspace - Convert global memory to workspace-specific
  • Global Memory Resources (5 new resources)
    • memory://global/recent?limit=50 - Recent global memories
    • memory://global/by-type/{type}?limit=50 - Global memories by context type
    • memory://global/by-tag/{tag}?limit=50 - Global memories by tag
    • memory://global/important?min=8&limit=50 - Important global memories
    • memory://global/search?q=query&limit=10 - Search global memories
    • All global resources throw helpful errors in isolated mode

Changed

  • Enhanced CreateMemory schema with optional is_global field (defaults to false)
  • Updated MemoryEntry schema with is_global and workspace_id fields
  • Updated searchMemories() to support three workspace modes with weighted results
  • Enhanced getMemory() to check both workspace and global storage
  • All retrieval methods respect workspace mode configuration

Technical

  • Added WorkspaceMode enum and getWorkspaceMode() helper
  • Added global Redis key helpers: globalMemory(), globalMemories(), globalByType(), etc.
  • Added ConvertToGlobalSchema and ConvertToWorkspaceSchema validation
  • Updated MemoryStore with convertToGlobal() and convertToWorkspace() methods
  • All global resources validate workspace mode and provide clear error messages
  • Backward compatible: defaults to isolated mode, no breaking changes

Use Cases

  • Personal preferences across all projects (coding standards, communication style)
  • Team conventions and organizational knowledge
  • Cross-project patterns and solutions
  • Shared tools and commands

Tools: 15 total (6 core + 3 smart context + 4 advanced + 2 global) Resources: 14 total (9 workspace + 5 global) Prompts: 1 (workspace_context)

See WORKSPACE_MODES.md for detailed documentation.

[1.2.1] - 2025-10-02

Added

  • Organizational Shared Memory Documentation - Comprehensive guide for team collaboration
    • Setup instructions for shared Redis instances
    • Security considerations for organizational deployments
    • Example workflows for cross-team learning
  • Improved Installation Instructions
    • Added Claude CLI installation method (npx @modelcontextprotocol/create-server)
    • Direct npx usage examples with no installation required
    • Multiple installation options (CLI, global, from source)

Changed

  • Enhanced README.md with organizational memory use cases
  • Updated installation section with recommended approaches
  • Clarified team collaboration scenarios

[1.2.0] - 2025-10-02

Added

  • TTL Support - Auto-expiring memories with ttl_seconds parameter (minimum 60s)
  • Export/Import Tools - Backup and restore memories
    • export_memories - Export to JSON with optional filtering by type/importance
    • import_memories - Import from JSON with embedding regeneration option
  • Duplicate Detection - Find and merge similar memories
    • find_duplicates - Detect duplicates with similarity threshold (default 0.85)
    • Auto-merge option with configurable keep strategy
  • Memory Consolidation - consolidate_memories - Manually merge multiple memories
  • Analytics Dashboard - memory://analytics resource
    • Memory trends (24h, 7d, 30d)
    • Top tags and importance distribution
    • Activity breakdown by day and type
  • Cloud Redis Documentation - Setup guides for Upstash, Redis Cloud, Railway

Changed

  • Resource URI handling - Fixed pathname parsing for memory:// scheme
  • Package renamed from @joseairosa/mcp-memory to @joseairosa/recall
  • Binary command: mcp-memory → recall

Technical

  • Added mergeMemories() method to MemoryStore
  • Added ConsolidateMemoriesSchema, ExportMemoriesSchema, ImportMemoriesSchema, FindDuplicatesSchema
  • Created src/tools/export-import-tools.ts
  • Created src/resources/analytics.ts

Tools: 13 total (6 core + 3 smart context + 4 advanced) Resources: 9 total Prompts: 1 (workspace_context)

[1.1.0] - 2025-10-02

Added

  • Smart Context Management
    • recall_relevant_context - Proactive memory retrieval based on current task
    • analyze_and_remember - AI-powered conversation analysis and extraction
    • summarize_session - Create session snapshots with summaries
  • Auto-Injection Prompt
    • workspace_context - Automatically injects critical directives and decisions at conversation start
  • Workspace Isolation - Memories automatically segmented by project directory
  • AI Analysis - Claude Haiku integration for intelligent memory extraction

Changed (Breaking)

  • API Migration: OpenAI → Anthropic Claude
    • Environment variable: OPENAI_API_KEY → ANTHROPIC_API_KEY
    • Embeddings: OpenAI 1536-dim → Hybrid (Claude keywords + trigrams) 128-dim
    • Cost reduction: ~$2/day → ~$0.20/day
  • Redis Key Structure: Added workspace prefix ws:{workspace_id}:
  • Package Dependencies: openai → @anthropic-ai/sdk

Technical

  • Added ConversationAnalyzer class for AI-powered analysis
  • Created src/analysis/conversation-analyzer.ts
  • Created src/prompts/ directory (index.ts, formatters.ts)
  • Created src/tools/context-tools.ts
  • Updated src/embeddings/generator.ts - Hybrid embedding approach
  • Added MCP prompts capability to server

Tools: 9 total (6 core + 3 smart context) Resources: 8 total Prompts: 1 (new capability)

[1.0.0] - 2025-10-02

Added

  • Initial Release - Persistent memory server for Claude using MCP protocol
  • 6 Core Tools
    • store_memory - Store single memory with metadata
    • store_batch_memories - Batch store multiple memories
    • update_memory - Modify existing memory
    • delete_memory - Remove memory by ID
    • search_memories - Semantic search using embeddings
    • organize_session - Create session snapshots
  • 8 Resources - Read-only memory access
    • memory://recent - Recent memories (default 50)
    • memory://by-type/{type} - Filter by context type
    • memory://by-tag/{tag} - Filter by tag
    • memory://important - High importance (≥8)
    • memory://session/{session_id} - Session memories
    • memory://sessions - All sessions list
    • memory://search?q=query - Search interface
    • memory://summary - Statistics overview
  • 10 Context Types
    • directive, information, heading, decision, code_pattern
    • requirement, error, todo, insight, preference
  • Features
    • Redis in-memory storage
    • OpenAI embeddings for semantic search
    • ULID identifiers
    • Importance scoring (1-10)
    • Tag-based organization
    • Session management
    • Zod schema validation

Technical

  • TypeScript with ESM modules
  • Redis (ioredis) for storage
  • OpenAI text-embedding-3-small (1536-dim)
  • MCP SDK v1.0.4
  • Multiple access patterns (hashes, sets, sorted sets)

Tools: 6 Resources: 8 Prompts: 0

Migration Guides

Migrating from v1.2.1 to v1.3.0

No breaking changes. All existing functionality works identically.

Defaults:

  • Workspace mode: isolated (same as v1.2.1)
  • All memories: is_global: false (workspace-only)
  • No configuration changes required

New features available:

  1. Enable global memories - Set WORKSPACE_MODE=hybrid in config
  2. Convert existing memories - Use convert_to_global tool
  3. Browse global memories - Use memory://global/* resources

Configuration example:

{
  "env": {
    "WORKSPACE_MODE": "hybrid",  // Enable both workspace + global
    "REDIS_URL": "redis://localhost:6379",
    "ANTHROPIC_API_KEY": "sk-ant-..."
  }
}

Use cases:

  • Keep mode as isolated for project-only memories (default behavior)
  • Use hybrid for personal preferences + project memories
  • Use global for team-wide shared knowledge across all projects

See WORKSPACE_MODES.md for detailed guide.

Migrating from v1.1.0 to v1.2.0

No breaking changes. All existing functionality works identically.

New features available immediately:

  • Set ttl_seconds when creating memories
  • Use export_memories and import_memories tools
  • Access memory://analytics resource

Optional: Upgrade Redis connection

  • Consider cloud Redis for remote access (Upstash, Redis Cloud, Railway)
  • See README for setup instructions

Migrating from v1.0.0 to v1.1.0

Breaking changes - Action required:

  1. Update environment variable
// Before
{
  "env": {
    "OPENAI_API_KEY": "sk-..."
  }
}

// After
{
  "env": {
    "ANTHROPIC_API_KEY": "sk-ant-..."
  }
}
  1. Embeddings migration (optional but recommended)
// Export without embeddings
{tool: "export_memories", args: {include_embeddings: false}}

// Import with regenerated embeddings
{tool: "import_memories", args: {data: "...", regenerate_embeddings: true}}
  1. Workspace awareness
  • Memories are now isolated by directory
  • Old memories without workspace prefix may need manual migration
  • Use Redis CLI to add ws:{id}: prefix if needed

Benefits:

  • 10x cost reduction (~$2/day → ~$0.20/day)
  • No OpenAI dependency
  • Faster embedding generation
  • Better context management with smart tools

Migrating to v1.0.0

Initial release - no migration needed.

Links