memcurve
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| analyze_for_recallA | Analyze message for recall/search intent. Args: message: User message text. Returns: Dict with: should_search, confidence, suggested_query, suggested_tags, suggested_entities, reasoning, phrase_signals. Raises: ValueError: Invalid input. |
| analyze_messageB | Analyze message for memory-worthy content. Args: message: User message text. Returns: Dict with: should_save, confidence, suggested_entities, suggested_tags, suggested_strength, reasoning, phrase_signals. Raises: ValueError: Invalid input. |
| auto_recall_process_messageC | Automatically recall and reinforce memories related to message topics. Args: message: User message to analyze (non-empty string). Returns: Dict with success, enabled, topics_found, memories_found, memories_reinforced, mode, and message. Raises: ValueError: If message is empty or invalid. |
| backfill_embeddingsA | Generate embeddings for memories that lack them. Args: model: Model name (default: all-MiniLM-L6-v2). limit: Max memories to process (1-10k, None=all). force: Regenerate existing embeddings. dry_run: Preview only. Returns: Dict with: success, processed, errors, model, total_memories, memories_without_embeddings, message. Raises: ValueError: Invalid limit. ImportError: sentence-transformers not installed. |
| cluster_memoriesA | Cluster similar memories or find duplicates. Args: strategy: Clustering strategy (default "similarity"). threshold: Similarity threshold (0.0-1.0, uses config default if None). max_cluster_size: Max cluster size (1-100, uses config default if None). find_duplicates: Find duplicate pairs instead of clustering. duplicate_threshold: Duplicate threshold (0.0-1.0, uses config default if None). Returns: Dict with clusters or duplicates list, scores, suggested_action. Raises: ValueError: Invalid threshold or max_cluster_size. |
| consolidate_memoriesA | Merge or link similar memories algorithmically. Args: cluster_id: Cluster UUID to act on (required unless auto_detect=True). mode: "preview" (no changes), "apply" (merge), or "link" (relations only). auto_detect: Find high-cohesion clusters automatically. cohesion_threshold: Minimum cohesion (0.0-1.0). Returns: Dict with success, mode, consolidated clusters, and results. Raises: ValueError: If cluster_id invalid or cohesion_threshold out of range. |
| create_relationA | Link two memories with typed relationship. Args: from_memory_id: Source memory UUID. to_memory_id: Target memory UUID. relation_type: Type (related, causes, supports, contradicts, has_decision, consolidated_from). strength: Relation strength (0.0-1.0). metadata: Optional metadata dict. Returns: Dict with relation_id, from, to, type, strength. Raises: ValueError: Invalid UUID or relation type. |
| gcA | Remove or archive low-scoring memories. Args: dry_run: Preview without removing. archive_instead: Archive instead of deleting. limit: Max memories to process (1-10,000). Returns: Dict with removed_count, archived_count, freed_score_sum, memory_ids. Raises: ValueError: Invalid limit range. |
| open_memoriesA | Retrieve specific memories by IDs with detailed information. Args: memory_ids: Single ID or list of IDs (max 100). include_relations: Include outgoing/incoming relations. include_scores: Include decay scores and age. page: Page number (1-indexed). page_size: Memories per page (max 100). Returns: Dict with memories, not_found IDs, and pagination metadata. Raises: ValueError: If memory ID invalid or list exceeds max length. |
| promote_memoryA | Promote high-value memories to long-term storage. Args: memory_id: Memory UUID to promote. auto_detect: Detect promotion candidates automatically. dry_run: Preview without promoting. target: Storage format ("obsidian"). force: Promote even if criteria not met. Returns: Dict with candidates, promoted_ids, and statistics. Raises: ValueError: If memory_id invalid or target unsupported. |
| read_graphA | Return entire knowledge graph with memories and relations. Args: status: "active", "promoted", "archived", or "all". include_scores: Include decay scores and age. limit: Max memories (1-10,000). page: Page number (1-indexed). page_size: Memories per page (max 100). Returns: Dict with memories, relations, stats, and pagination metadata. Raises: ValueError: If status invalid or limit out of range. |
| save_memoryA | Save memory to short-term storage with auto-enrichment. Args: content: Memory content (max 50k chars). tags: Tags (max 50). entities: Named entities (max 100, auto-extracted if None). source: Source (max 500 chars). context: Context (max 1k chars). meta: Custom metadata dict. strength: Base strength (1.0-2.0, auto-calculated if None). Returns: Dict with: success, memory_id, message, has_embedding, enrichment_applied. Raises: ValueError: Invalid input. |
| search_memoryA | Search memories with filters and pagination. Args: query: Search text (max 50k chars). tags: Filter by tags (max 50). status: Filter by status ('active', 'promoted', 'archived' or list of these). Defaults to ['active', 'promoted'] if None. top_k: Max results (1-100). window_days: Recent memories only (1-3650 days). min_score: Min decay score (0.0-1.0). use_embeddings: Enable semantic search. include_review_candidates: Include review-due memories. page: Page number (default: 1). page_size: Results per page (10-100, default: 10). preview_length: Content chars (0-5000, default: 300). Returns: Dict with results list and pagination metadata. Raises: ValueError: Invalid parameters. |
| search_unifiedA | Search across STM and LTM with unified ranking. Args: query: Search text (max 50k chars). tags: Filter by tags (max 50). status: Filter STM by status ('active', 'promoted', 'archived' or list of these). Defaults to ['active', 'promoted'] if None. limit: Max results (1-100). stm_weight: STM multiplier (0.0-2.0). ltm_weight: LTM multiplier (0.0-2.0). window_days: Recent STM only (1-3650 days). min_score: Min STM score (0.0-1.0). page: Page number (default: 1). page_size: Results per page (10-100, default: 10). preview_length: Content chars (0-5000, default: 300). Returns: Dict with results from both STM/LTM and pagination metadata. Raises: ValueError: Invalid parameters. |
| touch_memoryA | Reinforce memory to reset decay. Args: memory_id: Memory UUID. boost_strength: Increase base strength. Returns: Dict with old_score, new_score, use_count, strength. Raises: ValueError: Invalid UUID. |
| get_performance_metricsA | Get current performance metrics. Returns: Dict with: operation stats, counts, timings. |
| reset_performance_metricsB | Reset all performance metrics. Returns: Dict with: success, message. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 17 tools
Most tools target distinct actions: analyze_for_recall vs analyze_message are differentiated by intent, and cluster_memories vs consolidate_memories separate detection from action. However, search_memory and search_unified overlap in capability, and auto_recall_process_message could be confused with the analyze_* tools since it also starts with message analysis.
The majority follow a clear verb_noun pattern like save_memory, search_memory, promote_memory, and create_relation. Deviations include the abbreviated 'gc' and the awkward compound 'auto_recall_process_message', which break the otherwise predictable pattern.
At 17 tools, the server is slightly above the ideal 3-15 range but still reasonably scoped for a memory system that handles search, recall, consolidation, promotion, graph relationships, and maintenance. Each tool serves a plausible purpose, though a few analysis/search tools could be merged.
The toolset covers the core memory lifecycle well: save, retrieve, search, reinforce, promote, consolidate, cluster, relate, and garbage collect. Minor gaps exist such as no explicit edit-memory or delete-relation tool, but the provided operations are mostly sufficient for the perceived domain.