HyperspaceDB MCP Server
OfficialClick on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@HyperspaceDB MCP Server search my knowledge base for 'vector databases' using hybrid search"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
HyperspaceDB MCP Server
The HyperspaceDB MCP Server (mcp-hyperspacedb) exposes the complete low-level data plane, cognitive geometry engine, graph navigation, and administration interface of HyperspaceDB to Large Language Models (LLMs) via the Model Context Protocol (MCP).
Looking for drop-in conversational agent memory?
If you are setting up plug-and-play long-term memory for Claude Desktop, Cursor, Windsurf, or OpenWebUI, use mcp-hyperspace-memory instead (npx -y mcp-hyperspace-memory@latest).mcp-hyperspacedb is the full-featured database and geometry server designed for developers, data pipelines, and advanced cognitive agents requiring fine-grained control over vector spaces, collections, graph topologies, and cache systems.
π Key Architectural Capabilities
1. Geometric Diagnostics & Curvature Analysis
Diagnose latent data topology using Gromov $\delta$-hyperbolicity over Riemannian and pseudo-Riemannian manifolds:
hyperspace_analyze_geometry: Employs the 4-point Gromov condition to recommend the ideal manifold metric (lorentz,poincare,cosine, orl2).
2. Cognitive Agent Diagnostics (Chain-of-Thought Stability)
Evaluate the geometric stability of multi-step reasoning traces:
hyperspace_analyze_thought_stability: Calculates Lyapunov Convergence of a reasoning trajectory. Distinguishes whether the model's reasoning converges toward a stable attractor or diverges into chaotic hallucinations.hyperspace_predict_momentum: Forecasts future thought trajectories via Koopman momentum extrapolation.hyperspace_get_trust_score: Computes geodesic trust scores along thought sequences.
3. Hyperbolic Hierarchy & Graph Traversal
hyperspace_get_subsumption_tree: Traces taxonomic subsumption trees in hyperbolic space ($H^{33}$).hyperspace_get_concept_parents: Identifies hypernym/parent concepts along the Lorentz cone.hyperspace_graph_traverse: Multi-hop BFS/DFS path exploration in the HNSW knowledge graph.hyperspace_explore_graph: Extracts interactive subgraphs (nodes and links) for 2D/3D visualization.hyperspace_find_clusters: Detects emergent semantic clusters directly in the vector space.
4. Advanced Search & Vector Data Plane
hyperspace_search_text: Hybrid semantic + BM25 lexical search with adjustable fusion weight (hybrid_alpha).hyperspace_search_wasserstein: Optimal Transport (Earth Mover's Distance) semantic distribution search.hyperspace_insert_text/hyperspace_get_points/hyperspace_delete_points: Point-level CRUD operations.
5. Collection Lifecycle & Hardware Acceleration Maintenance
Full control over collection creation with quantization schemes (
none,medium,medium_plus,turbo,extreme).L0 Hot Tier cache administration (
cache_stats,cache_clear,cache_config).Server-side flow matching reconsolidation (
trigger_reconsolidation), collection freezing, HNSW index rebuilding, and disk vacuuming.
Related MCP server: io.github.avaazquezz/mcp-qdrant
π οΈ Installation & MCP Host Configuration
Quick Start with npx
npx -y mcp-hyperspacedb@latestHost Configurations
Add to your MCP host configuration (e.g. claude_desktop_config.json, ~/.codeium/windsurf/mcp_config.json, or Cursor Settings):
βοΈ 1. YAR.INK Managed SaaS
{
"mcpServers": {
"hyperspacedb": {
"command": "npx",
"args": ["-y", "mcp-hyperspacedb@latest"],
"env": {
"HYPERSPACE_HOST": "the.yar.ink",
"HYPERSPACE_API_KEY": "sk_YOUR_API_KEY"
}
}
}
}π 2. Local Self-Hosted Instance (Default Auth)
{
"mcpServers": {
"hyperspacedb": {
"command": "npx",
"args": ["-y", "mcp-hyperspacedb@latest"],
"env": {
"HYPERSPACE_HOST": "localhost:50051",
"HYPERSPACE_API_KEY": "I_LOVE_HYPERSPACEDB"
}
}
}
}π 3. Local DB + Cloud v5 Embedding API Key (Two-Key Setup)
When running a secured local engine while leveraging continuous cloud embeddings:
{
"mcpServers": {
"hyperspacedb": {
"command": "npx",
"args": ["-y", "mcp-hyperspacedb@latest"],
"env": {
"HYPERSPACE_HOST": "localhost:50051",
"HYPERSPACE_API_KEY": "my_local_secret",
"YAR_API_KEY": "sk_YOUR_YAR_API_KEY"
}
}
}
}HYPERSPACE_API_KEY: Authenticates with local gRPC engine (localhost:50051).YAR_API_KEY(or legacyCDE_API_KEY): Authenticates with cloud embeddings endpoint.
π§© Complete Tool Inventory (31 Tools)
π Data Plane Tools
Tool | Required Parameters | Optional Parameters | Description & Return Value |
|
|
| Semantic + lexical hybrid search. Returns array of matching records with |
|
|
| Optimal Transport (Wasserstein) search. Returns array of top matches ranked by Wasserstein distance. |
|
|
| Inserts text under numeric |
|
| β | Retrieves vector coordinates and metadata for a list of point IDs. Returns JSON array of points. |
|
|
| Creates a new vector space with chosen geometry, quantization, and optional client-side Zero-Knowledge encryption. Returns |
|
| β | Permanently deletes a collection and its vectors. Returns |
| β | β | Lists all active collections. Returns array of |
πΈοΈ Graph & Hierarchical Navigation Tools
Tool | Required Parameters | Optional Parameters | Description & Return Value |
|
|
| Explicitly creates a directed, labeled, weighted edge between two entities in the CSR vector-graph. |
|
| β | Atomically bulk-inserts explicit directed edges in the CSR graph. |
|
|
| Deletes an explicit directed edge with zero-lag invalidation. |
|
| β | Declaratively induces graph edges from existing document metadata ( |
|
|
| Multi-hop BFS graph traversal starting from |
|
|
| Traverses graph and outputs nodes and links formatted for visual renderers. |
|
|
| Returns local connectivity and neighboring nodes at a given HNSW graph layer. |
|
|
| Returns hierarchical Lorentz cone subsumption tree starting from |
|
|
| Traverses upward along the Lorentz cone to retrieve parent concepts. |
|
|
| Detects emergent semantic clusters in vector space. Returns cluster centroids and member IDs. |
π§ Geometry & Cognitive AI Tools
Tool | Required Parameters | Optional Parameters | Description & Return Value |
|
|
| Computes Gromov $\delta$-hyperbolicity. Returns |
|
|
| Computes Lyapunov exponent along a Chain-of-Thought path. Classifies as |
|
|
| Forecasts next cognitive steps using Koopman momentum extrapolation. |
|
|
| Evaluates geodesic path consistency and returns trust score (0.0β1.0). |
|
|
| AI Sleep Mode: Triggers Flow Matching server optimization for concepts in collection. |
βοΈ System, Cache & Maintenance Tools
Tool | Required Parameters | Optional Parameters | Description & Return Value |
|
| β | Merges gRPC telemetry, collection stats, logical clock, and HTTP cache stats into a JSON object. |
|
| β | Retrieves hit rate, miss rate, and memory usage for the L0 Hot Tier Cache. |
|
| β | Flushes all entries from the L0 Cache for the target collection. Returns |
|
|
| Updates L0 Cache eviction policy ( |
|
| β | Locks collection into read-only mode to prevent new inserts. |
|
| β | Unlocks a frozen collection to permit modifications. |
|
| β | Triggers server-side HNSW index rebuild and optimization. |
| β | β | Purges deleted tombstones and reclaims fragmented disk space. |
π» Local Development
cd integrations/mcp-hyperspacedb
npm install
npm run build
npm run devπ License
MIT Β© YARlabs
Available Tools
35 toolsclarify_and_split_intentB
Alias for hyperspace_query_planner. Evaluates intent clarity, asks clarifying questions if ambiguous, and splits multi-part questions into isolated search queries.
Inputs: raw_user_query (required), context, collections, max_sub_queries.
hyperspace_analyze_geometryA
Calculate Gromov Delta-hyperbolicity either for a named collection on the server or over supplied vectors (array of numeric arrays). Returns {delta, recommendation: 'lorentz'|'poincare'|'cosine'|'l2'}.
Inputs: samples, vectors, collection.
hyperspace_analyze_thought_stabilityA
Calculate Lyapunov Convergence of a trajectory (Chain of Thought). Returns a JSON object {lyapunov_exponent, classification: 'stable'|'chaotic'} - negative result means stable/converging, positive means chaotic/hallucinating.
Inputs: trajectory (required), curvature.
hyperspace_cache_clearB
Clear / purge all items in the L0 Cache for a specific collection. Returns ok/error status object.
Inputs: collection (required).
hyperspace_cache_configB
Update L0 Cache configuration (eviction policy, ANN threshold) for a specific collection. Returns updated config as JSON result.
Inputs: policy (required), collection (required), ann_threshold.
hyperspace_cache_statsB
Get cache statistics (hits, misses, policy, etc.) for a specific collection's L0 Hot Tier Cache. Returns stats object as JSON result.
Inputs: collection (required).
hyperspace_create_collectionC
Setup new memory spaces (collections) with specific vector dimension, metric geometry, and quantization level. Returns the created collection config as JSON result.
Inputs: collection (required), metric, dimension, quantization, +1 more.
hyperspace_delete_collectionB
Permanently delete a collection and all of its vectors. Returns a result object with ok status or error.
Inputs: collection (required).
hyperspace_delete_pointsC
Delete a single vector point from a collection by its ID. Returns a result object with ok status or error.
Inputs: id (required), collection (required).
hyperspace_explore_graphB
Traverse the graph and return nodes and links in a format ready for visualization. Requires collection and start_id of the starting node.
Inputs: start_id (required), collection (required), max_depth, max_nodes.
hyperspace_find_clustersA
Detect emergent structure/hierarchy in collection. Returns a JSON array of clusters, each {cluster_id, members: [...], centroid, size}, filtered by min_cluster_size.
Inputs: collection (required), min_cluster_size.
hyperspace_freeze_collectionB
Freeze a collection to make it read-only, preventing new inserts. Returns ok/error status object.
Inputs: collection (required).
hyperspace_get_concept_parentsB
Retrieve parent concepts in a hierarchical collection. Requires collection and the concept id. Returns a JSON list of parent nodes.
Inputs: id (required), collection (required), layer, limit.
hyperspace_get_neighborsB
Explore local connectivity of a concept in the vector index graph. Requires collection name and id of the node/concept. Returns a list of neighboring nodes as JSON result.
Inputs: id (required), collection (required), layer, limit.
hyperspace_get_pointsB
Retrieve vector coordinate and metadata for a list of point IDs. Requires collection name and ids array of point IDs.
Inputs: ids (required), collection (required).
hyperspace_get_statsB
Get detailed statistics for the named collection. Returns a JSON object with count, dimension, metric, logical_clock, and storage stats. Merges metadata, cache stats and WriteBuffer size.
Inputs: collection (required).
hyperspace_get_subsumption_treeB
Retrieve the Lorentz hierarchy subsumption tree starting from a given root_id; returns the tree as JSON in the given collection.
Inputs: root_id (required), collection (required), max_depth.
hyperspace_get_trust_scoreC
Evaluate stability and trust score for a given thought trajectory path. Requires collection and trajectory_ids array of trajectory node IDs.
Inputs: collection (required), trajectory_ids (required), curvature.
hyperspace_graph_traverseA
Deep graph exploration starting from node start_id inside collection. Finds logical paths between concept A and context B for complex reasoning/cross-referencing. Returns a JSON object {paths: [...], nodes: [...], edges: [...]} bounded by max_depth and max_nodes.
Inputs: start_id (required), collection (required), max_depth, max_nodes.
hyperspace_induce_edgesC
Declaratively induce explicit graph edges from existing document metadata attributes (path_hierarchy, foreign_key, exact_match). Parameters: collection, rules array.
Inputs: rules (required), collection (required).
hyperspace_insert_textB
Store a new factual claim or memory: insert text into collection under numeric id (auto-vectorized). Returns a JSON object {ok: true, id} on success or {ok: false, error} on failure.
Inputs: id (required), text (required), collection (required), metadata.
hyperspace_linkB
Create an explicit directed, labeled, weighted relationship edge between two entities in the native CSR vector-graph. Parameters: collection, source_id, target_id, optional relation (default 'OWNS'), weight (default 1.0).
Inputs: source_id (required), target_id (required), collection (required), weight, +1 more.
hyperspace_link_batchB
Bulk-insert explicit directed edges in CSR graph atomically. Parameters: collection, edges array of {source_id, target_id, relation, weight}.
Inputs: edges (required), collection (required).
hyperspace_list_collectionsA
List all active collections. Returns a JSON array of objects, each with {name, dimension, metric, count} metadata fields.
Inputs: no named parameters.
hyperspace_pdk_extensionB
Invoke a native HyperspaceDB PDK Extension (e.g. 'git_ast' for AST codebase indexing & blast radius, 'embed' for structure-aware Markdown ingestion, 'shield', 'chem', 'bio', 'fin', 'sec', 'legal', 'onto').
Inputs: action (required), extension (required), payload.
hyperspace_predict_momentumB
Forecast future agent thought paths using Koopman momentum extrapolation. Requires collection and trajectory_ids array; steps sets prediction horizon. Returns predicted trajectory summary as JSON.
Inputs: collection (required), trajectory_ids (required), steps, curvature.
hyperspace_query_plannerA
Enterprise RAG Query Planner. Evaluates user intent before vector search: detects ambiguous/underspecified questions (requesting multiple-choice user clarification) and decomposes multi-part queries into parallel sub-queries.
Inputs: raw_user_query (required), context, collections, max_sub_queries.
hyperspace_rebuild_indexB
Rebuild and optimize the HNSW index on the server for a specific collection. Returns rebuild status/result as JSON.
Inputs: collection (required).
hyperspace_search_textB
Search a collection for entries semantically similar to text (natural language query). Returns a JSON array of up to top_k results, each with id, score, and stored payload.
Inputs: text (required), collection (required), top_k, bm25_options, +1 more.
hyperspace_search_wassersteinA
Advanced Optimal Transport (Wasserstein) search inside collection for text. Compares distributions/complex concept overlap. Returns a JSON array of top_k matches ranked by Wasserstein distance, each with id, score, payload.
Inputs: text (required), collection (required), top_k.
hyperspace_trigger_reconsolidationB
AI Sleep Mode: trigger Flow Matching on the server to optimize the geometric representation of concepts in collection based on usage/context. Returns a JSON object {ok, iterations, loss_before, loss_after}.
Inputs: collection (required), learning_rate.
hyperspace_unfreeze_collectionA
Unfreeze a previously frozen collection to allow inserts again. Returns ok/error status object.
Inputs: collection (required).
hyperspace_unlinkC
Delete an explicit directed relationship edge with zero-lag invalidation. Parameters: collection, source_id, target_id, optional relation.
Inputs: source_id (required), target_id (required), collection (required), relation.
hyperspace_vacuumB
Perform vacuuming on the database to permanently purge deleted vectors and reclaim disk space. Returns vacuum status and reclaimed-space result.
Inputs: no named parameters.
hyperspace_verify_generationC
Tribunal Router (Mathematical Anti-Hallucination Guardrail): Verify whether an LLM generated answer stays within the geometric entailment cones and geodesic boundaries of retrieved context IDs.
Inputs: collection (required), llm_answer (required), context_ids (required), check_geodesic, +1 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelogβ¦
TDQS
Scored across 35 tools
Most tools are distinct, but there is direct duplication between hyperspace_query_planner and clarify_and_split_intent, which is explicitly an alias. Graph exploration tools (explore_graph, graph_traverse, get_neighbors, find_clusters) also overlap enough to cause misselection, despite descriptions helping.
Almost all tools follow a clear hyperspace_verb_noun snake_case pattern. The only notable deviation is clarify_and_split_intent, which lacks the server prefix and is an alias for another tool.
35 tools is heavy for the apparent scope, with many niche analysis, cache, and PDK extension operations. Several advanced tools feel bolted on rather than essential to the core vector-graph/memory workflow.
Coverage is broad: collections, points, text search, explicit and induced graph edges, traversal, clustering, cache, maintenance, and verification. Minor gaps exist around updating existing point content and updating collection configuration, but most workflows have a path.
Maintenance
Related MCP Connectors
MCP facade over the Nebelus Construction API. ~48 tools give full agent build parity: create/update/probe agents, edit graphs, attach knowledge and vector stores, wire connectors, set governance policies and locked guardrails, enable grounding-trace, and read deployment wiring. Purpose-built for regulated industries: data residency is enforced per region (EU / GCC-KSA), with PII controls and an audit trail. Agents are created as drafts β no deploy tool is exposed over MCP by design; publishing happens in the Nebelus console.
Governed data discovery, exact queries, decisions, simulations, and runtime utilities over MCP.
214 tools for AI agents over one MCP endpoint: image, video, audio, 3D, data, storage, mail, GPUs.
MCP-Native LLM Orchestration Agent
Related MCP Servers
- AlicenseAqualityDmaintenanceEnables MCP-capable agents to create and invoke OpenServerless endpoints, manage secrets, and configure integrations such as S3, PostgreSQL, Redis, Milvus, and MongoDB.13Apache 2.0
- AlicenseAqualityBmaintenanceEnables AI assistants to interact with a Qdrant vector database by exposing collection, point, vector, payload, snapshot, search, recommendation, discovery, and observability operations as MCP tools.13168 PyPI2MIT
- AlicenseNot gradedqualityBmaintenanceEnables agents to turn documents into vectors, perform approximate nearest neighbor search with HNSW, and filter by metadata via MCP tools for RAG workflows.MIT
- FlicenseBqualityBmaintenanceEnables AI agents to query an air-gapped enterprise knowledge vault via seven MCP tools for deterministic and hybrid retrieval, entity graph traversal, context compression, archive inspection, onboarding scans, and node status.7-