Create new entities in the knowledge graph.

Each entity needs: name (str), entityType (str). Optional: observations (list[str]), tags (list[str]), metadata (dict). Skips duplicates by name.

Create relations between entities.

Each relation needs: from (str), to (str), relationType (str). Optional: weight (float), tags (list[str]). Both entities must exist. Use active voice for relationType.

Add observations to existing entities.

Each item needs: entityName (str), contents (list[str]). Optional: source (str), confidence (float 0-1), tags (list[str]), importance (float 0-1, default 0.5 — higher survives compression), obs_type (str: fact, decision, preference, api_endpoint, dependency, file_path, code_snippet, config, schema). Deduplicates by exact match and semantic similarity. Protected obs_types (api_endpoint, dependency, file_path, code_snippet, config, schema) always survive compression.

Delete entities (soft delete by default, cascades to relations).

Set hard=true for permanent deletion.

Delete specific observations from entities.

Each item needs: entityName (str), observations (list[str] of content to delete).

Delete relations from the knowledge graph.

Each item needs: from (str), to (str), relationType (str).

Read the knowledge graph (compressed by default).

Filter by tags, entity_types. Set compress=false for full JSON.

Full-text search with BM25 ranking.

Supports prefix search, phrases ("exact match"), boolean (AND/OR/NOT). Filter by tags, entity_types, time_range ([start, end] ISO strings).

Open specific entities by name.

depth=0: exact entities only. depth=1: include direct neighbors via relations. depth=2+: BFS expansion.

Lightweight context snapshot (~200-500 tokens) for scoped durable recall.

Returns: pinned entities, recent activity, hint-matched entities, and graph stats. Call when prior sessions, stable project facts, or cross-session continuity may matter. Skip for one-off answers or tasks already fully grounded in the current context. Pass hint='current topic' to get relevant entities surfaced. Pass project='name' to scope results to a specific project tag. Pass limit to control how many hint matches are included; prefer 3-5 unless deeper recall is justified.

Record what happened this turn. Auto-creates missing entities.

Common actions: file_changed, decision_made, bug_fixed, feature_added, refactored, investigated, discussed, preference_set.

Query activity timeline.

time_range: relative like "2h", "7d", "30m". Or use start/end with ISO datetime strings. Filter by actions, entity_name, session_id.

Manage tags. Actions: list, create, delete, tag, untag, cleanup.

list: show all tags. create: new tag (name required, optional description/color/auto_expire_hours). delete: remove a user tag (cannot delete system tags). tag: apply tag_name to entity_name. untag: remove tag_name from entity_name. cleanup: remove expired ephemeral items.

Merge source entity into target. Source is soft-deleted.

strategy: 'combine' (move all observations) or 'dedupe' (skip duplicates). Relations and tags are transferred. Self-relations are removed.

Export the full knowledge graph.

format: 'json' or 'jsonl' (compatible with old @modelcontextprotocol/server-memory).

Import knowledge graph data. Auto-detects JSON or JSONL format.

Compatible with old @modelcontextprotocol/server-memory JSONL files. Skips duplicate entities. Skips relations to missing entities.

Get memory statistics: entity/relation/observation counts, tag distribution, DB size, orphan entities, deleted items.

Create a backup of the memory database.

Default destination: ~/.local/share/server-memory/backups/memory-YYYYMMDD-HHMMSS.db Provide dest_path to override the backup location.

Get observation version history for an entity.

Shows current content, version number, importance, obs_type, and all previous versions with timestamps. Use content_prefix to filter to specific observations.

Rich context snapshot for rare deep bootstrap (~500-1500 tokens).

Returns all pinned entities with full observations, recent activity (last 10), and recently changed entities. Use only when compact recall is insufficient for a cross-session task. Prefer memory_context for ordinary scoped recall.