spatial-memory-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| SPATIAL_MEMORY_LOG_LEVEL | No | Logging verbosity | INFO |
| SPATIAL_MEMORY_MEMORY_PATH | No | LanceDB storage directory | ./.spatial-memory |
| SPATIAL_MEMORY_OPENAI_API_KEY | No | Required only for OpenAI embeddings | |
| SPATIAL_MEMORY_EMBEDDING_MODEL | No | Embedding model (or openai:text-embedding-3-small) | all-MiniLM-L6-v2 |
| SPATIAL_MEMORY_EMBEDDING_BACKEND | No | auto (ONNX if available), onnx, or pytorch | auto |
| SPATIAL_MEMORY_AUTO_DECAY_ENABLED | No | Automatic importance decay over time | true |
| SPATIAL_MEMORY_COGNITIVE_OFFLOADING_ENABLED | No | Enable queue-based auto-capture pipeline | false |
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
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| rememberC | Store a new memory in the spatial memory system. |
| remember_batchB | Store multiple memories efficiently in a single operation. |
| recallB | Search for similar memories using semantic similarity. |
| nearbyC | Find memories similar to a specific memory. |
| forgetA | Delete a memory by its ID. |
| forget_batchA | Delete multiple memories by their IDs. |
| healthC | Check system health status. |
| journeyB | Navigate semantic space between two memories using spherical interpolation (SLERP). Discovers memories along the conceptual path. |
| wanderA | Explore memory space through random walk. Uses temperature-based selection to balance exploration and exploitation. |
| regionsA | Discover semantic clusters in memory space using HDBSCAN. Returns cluster info with representative memories and keywords. |
| visualizeB | Project memories to 2D/3D for visualization using UMAP. Returns coordinates and optional similarity edges. |
| decayA | Apply time and access-based decay to memory importance scores. Memories not accessed recently will have reduced importance. |
| reinforceA | Boost memory importance based on usage or explicit feedback. Reinforcement increases importance and can reset decay timer. |
| extractC | Automatically extract memories from conversation text. Uses pattern matching to identify facts, decisions, and key information. |
| consolidateA | Merge similar or duplicate memories to reduce redundancy. Finds memories above similarity threshold and merges them. |
| statsB | Get database statistics and health metrics. |
| namespacesA | List all namespaces with memory counts and date ranges. |
| delete_namespaceA | Delete all memories in a namespace. DESTRUCTIVE - use dry_run first. |
| rename_namespaceA | Rename a namespace, moving all its memories to the new name. |
| export_memoriesB | Export memories to file (Parquet, JSON, or CSV format). |
| import_memoriesB | Import memories from file with validation. Use dry_run=true first. |
| hybrid_recallA | Search memories using combined vector and keyword (full-text) search. |
| setup_hooksB | Generate hook configuration for cognitive offloading. Returns ready-to-use hooks JSON for Claude Code or Cursor. |
| discover_connectionsA | Find cross-corpus connections for a memory. Discovers semantically similar memories across all namespaces and projects using ANN-based search with pluggable scoring. |
| corpus_bridgesA | Find cross-namespace bridges in the memory corpus. Discovers memories in different namespaces that are semantically similar -- potential knowledge links or duplicates. |
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 25 tools
Most tools target distinct operations, but there is notable overlap among retrieval/exploration tools: recall, hybrid_recall, nearby, discover_connections, corpus_bridges, journey, and wander all return semantically related memories with only subtle differences in scope. The descriptions help distinguish them, but an agent could easily select the wrong one, especially for recall vs. hybrid_recall and discover_connections vs. corpus_bridges.
Names are consistently lowercase snake_caseaine and readable, but they do not follow a single verb_noun convention; many are bare verbs (forget, recall, reinforce), while others are nouns or noun phrases (regions, stats, namespaces, corpus_bridges). This mix of imperative actions and declarative nouns is predictable enough to navigate but lacks the tight pattern of the highest-calibration servers.
At 25 tools, the server sits at the heavy end of the borderline range; the spatial memory domain can justify many operations, but the count feels inflated by a cluster of overlapping search and exploration tools. A tighter set closer to 18-20 tools would likely be just as capable and easier for an agent to navigate.
The surface covers the core memory lifecycle well: create (remember, remember_batch, extract), read/retrieve (recall, hybrid_recall, nearby), delete (forget, forget_batch, delete_namespace), plus namespace management, import/export, analytics, and maintenance. The main gap is the lack of a direct get_memory-by-ID tool and an update tool for editing memory content; users must work around this with forget-and-remember or importance-only mutation tools.