hyperspacedb-mcp
GitHub提供HyperspaceDB的MCP服务器配置与使用指南,涵盖数据库操作、向量CRUD、语义搜索及图遍历等30+工具。支持Claude Desktop、Cursor等AI宿主的环境变量设置与JSON配置方法,助力开发者集成高性能向量数据库能力。
触发场景
安装
npx skills add YARlabs/hyperspace-db --skill hyperspacedb-mcp -g -y
SKILL.md
Frontmatter
{
"name": "hyperspacedb-mcp",
"description": "Model Context Protocol (MCP) server for HyperspaceDB. Use this skill to configure, connect, and use HyperspaceDB tools in Claude Desktop, Cursor, Windsurf, Antigravity, or any MCP-compatible AI host. Covers all 30+ available MCP tools. Trigger on: \"MCP\", \"Claude Desktop\", \"Cursor MCP\", \"hyperspace_search\", \"hyperspace_insert\", \"model context protocol\", \"mcp-hyperspacedb\", \"tool use\"."
}
HyperspaceDB MCP Servers
HyperspaceDB provides two official Model Context Protocol (MCP) servers:
mcp-hyperspacedb(Database Plane, v4.0.0): 27 tools for collection DDL, vector CRUD, HNSW graph traversal, Lyapunov stability analysis, Gromov delta hyperbolicity, and Koopman momentum.mcp-hyperspace-memory(Cognitive Memory Plane, v4.0.0): 9 tools for autonomous agent memory, episodic facts, session isolation, Fréchet mean consolidation, and hallucination verification.
Setup & Configuration
Add to your MCP config file (e.g., claude_desktop_config.json, .cursor/mcp.json):
{
"mcpServers": {
"hyperspacedb": {
"command": "npx",
"args": ["-y", "mcp-hyperspacedb"],
"env": {
"HYPERSPACE_HOST": "the.yar.ink",
"HYPERSPACE_API_KEY": "YOUR_YARINK_API_KEY"
}
},
"hyperspace-memory": {
"command": "npx",
"args": ["-y", "mcp-hyperspace-memory"],
"env": {
"HYPERSPACE_HOST": "the.yar.ink",
"HYPERSPACE_API_KEY": "YOUR_YARINK_API_KEY"
}
}
}
}
Config file locations by host:
| Host | Config path |
|---|---|
| Claude Desktop (macOS) | ~/Library/Application Support/Claude/claude_desktop_config.json |
| Claude Desktop (Windows) | %APPDATA%\Claude\claude_desktop_config.json |
| Cursor | .cursor/mcp.json in project root |
| Windsurf | ~/.windsurf/mcp.json |
| Antigravity / custom | Per host documentation |
Environment Variables
| Variable | Default | Description |
|---|---|---|
HYPERSPACE_HOST |
localhost:50051 |
gRPC address of HyperspaceDB node |
HYPERSPACE_API_KEY |
I_LOVE_HYPERSPACEDB |
Authentication key |
Never hardcode these values. Always pass via environment variables.
Complete Tool Reference
📁 Collection Management
| Tool | Description |
|---|---|
hyperspace_list_collections |
List all collections with stats |
hyperspace_create_collection |
Create a new collection (name, dimension, metric: cosine/l2/lorentz/poincare/hybrid, cascadePipeline) |
hyperspace_delete_collection |
Permanently delete a collection and all vectors |
hyperspace_freeze_collection |
Make collection read-only |
hyperspace_unfreeze_collection |
Re-enable inserts |
hyperspace_rebuild_index |
Rebuild and optimize the HNSW index |
hyperspace_vacuum |
Purge soft-deleted vectors, reclaim disk space |
📥 Data Operations
| Tool | Description |
|---|---|
hyperspace_insert_text |
Insert text (auto-embedded server-side) |
hyperspace_delete_points |
Delete a vector by ID |
hyperspace_get_points |
Retrieve vectors by IDs |
🔍 Search
| Tool | Description |
|---|---|
hyperspace_search_text |
Semantic search by text query; supports hybrid_alpha for BM25 fusion |
hyperspace_search_wasserstein |
Optimal Transport cross-distribution search |
🕸️ Graph & Hierarchy
| Tool | Description |
|---|---|
hyperspace_get_neighbors |
Direct HNSW neighbors of a node |
hyperspace_graph_traverse |
BFS/DFS multi-hop traversal |
hyperspace_explore_graph |
Visualization-ready graph data (nodes + edges) |
hyperspace_get_subsumption_tree |
Lorentz hierarchy tree from a root concept |
hyperspace_get_concept_parents |
Parent concepts in the hierarchy |
hyperspace_find_clusters |
Unsupervised cluster detection |
🧠 Cognitive AI
| Tool | Description |
|---|---|
hyperspace_analyze_thought_stability |
Lyapunov CoT convergence analysis (returns { lyapunov_exponent, is_stable }) |
hyperspace_predict_momentum |
Koopman trajectory momentum forecast (returns number[] — next vector) |
hyperspace_get_trust_score |
Composite reasoning trust score (returns number 0–1) |
hyperspace_analyze_geometry |
Gromov Delta analysis → optimal metric (cosine/l2/lorentz/poincare/hybrid) |
⚙️ System & Cache
| Tool | Description |
|---|---|
hyperspace_get_stats |
Node telemetry, vector counts, clock |
hyperspace_trigger_reconsolidation |
Trigger Flow Matching sleep-mode optimization |
hyperspace_cache_stats |
L0 cache hit/miss statistics |
hyperspace_cache_clear |
Clear L0 cache for a collection |
hyperspace_cache_config |
Update cache eviction policy and ANN threshold |
Example Prompts
Once connected, you can say to your AI agent:
"Create a collection called 'team_memory' with cosine metric and 1536 dimensions."
"Search 'team_memory' for documents about Kubernetes deployments."
"Analyze the geometry of these vectors and tell me the best metric to use."
"Check if my reasoning chain [ids: 1,2,3,4,5] is stable or diverging."
"Show me the parent concepts of node 42 in 'knowledge_graph'."
"What are the main topic clusters in 'research_papers'?"
How Cognitive Tools Work
The MCP server implements cognitive tools as client-side computations:
getPoints()fetches the stored vectors for the given trajectory IDs- Mathematical analysis (Lyapunov, Koopman, trust) runs inside the MCP server process
- Results are returned without any server-side RPC for the math itself
This means cognitive tools work against any version of the HyperspaceDB server.
Geometry awareness: When collection metric is hybrid, momentum extrapolation
splits the vector into Lorentz (first 33 dims) and Euclidean (remaining dims) parts,
extrapolates each independently, then recombines.
See Also
- hyperspacedb-core — direct SDK usage
- hyperspacedb-cognitive — cognitive primitives explained
- hyperspacedb-depin — DePIN node setup
版本历史
-
b520a65
当前 2026-10-19 11:05
memory server版本从v1.0.0升级至v4.0.0,新增Fréchet mean consolidation工具及内存压力测试套件
- c226b54 2026-09-09 10:13
- cc17a8d 2026-07-25 09:06


