NEURONMCP/MCP Server
NeuronMCP
Model Context Protocol server with 100+ tools (27 PostgreSQL + 70+ NeuronDB) enabling MCP-compatible clients (like Claude Desktop) to access NeuronDB. Vector operations, ML training, RAG pipeline, reranking, dataset loading, and PostgreSQL administration through JSON-RPC 2.0.
bash
Quickstart
./neurondb-mcp --help100+ Tools27 PostgreSQL ToolsJSON-RPC 2.0stdio Transport
NeuronMCP
MCP Server
Architecture
MCP server architecture with protocol handling, tools, resources, and middleware
NeuronMCP Architecture
MCP Client
Claude Desktop | Custom Clients | stdio Communication
Protocol Handler
- • JSON-RPC 2.0
- • stdio Transport
- • Request Routing
- • Error Handling
Tools
- • Vector Operations
- • ML Operations
- • Analytics
- • RAG Operations
Resources
- • Schema Info
- • Model Catalog
- • Index Config
- • Stats & Config
Middleware
Validation
Input checks
Logging
Structured logs
Timeout
Request limits
Error Handling
Graceful errors
NeuronDB PostgreSQL
Vector Search | ML Algorithms | Embeddings | SQL Functions
NeuronMCP Tools
100+ tools for vector ops, ML, RAG, and database admin
NeuronMCP
console • demo
v1.0ready
100+ MCP tools
Vector search, embeddings, and indexing tools
Vector Toolsdemo
// Call vector search tool
{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "vector_search",
"arguments": {
"table": "embeddings",
"query_vector": [0.1, 0.2, ...],
"top_k": 10,
"distance": "cosine"
}
}
}Results5 rows
| id | similarity | text |
|---|---|---|
| 42 | 0.9523 | Vector search result: semantic similarity matching in high-dimensional spaces… |
| 38 | 0.9234 | HNSW index result: fast approximate nearest neighbor search… |
| 35 | 0.8945 | Embedding generation result: text converted to numerical vectors… |
| 31 | 0.8656 | Index creation result: HNSW index built successfully… |
| 28 | 0.8367 | Similarity computation result: cosine distance calculated… |
Performance
Query Time
8.42ms
Latency (P95)
12.5ms
QPS
8.2k
Status
ready
Query Statistics
Execution
Rows Returned5
Cache Hit96%
PlanOptimized
Connection
Databaseneuronmcp
Versionv1.0
Statusactive
Summary
Total Queries1,247
Success Rate99.8%
Avg Latency8.2ms
MCP Server Features
Why NeuronMCP
MCP Protocol Implementation
- •Full JSON-RPC 2.0 implementation with stdio, HTTP, and SSE transport
- •Supports tools, resources, prompts protocol, sampling/completions, and progress tracking
- •Batch operations with transactional tool calls
- •Protocol version negotiation and capability discovery
- •Compatible with all MCP-compatible clients including Claude Desktop
Vector Operations Tools
- •Vector search with multiple metrics: L2, Cosine, Inner Product, Manhattan, Hamming, Jaccard
- •Similarity computation and embedding generation (text, image, multimodal)
- •Index creation: HNSW and IVF indexing
- •Batch embedding generation with intelligent caching
- •Hybrid search with reciprocal rank fusion
- •Multi-vector, faceted, temporal, and diverse search capabilities
ML Operations Tools
- •Complete ML pipeline: training, prediction, evaluation, and AutoML for all 52 algorithms
- •Model catalog management with versioning and A/B testing
- •Hyperparameter tuning support
- •ONNX model import, export, and inference
- •Model deployment workflows and monitoring
Reranking & RAG Operations
- •Multiple reranking strategies: cross-encoder, LLM-powered, Cohere, ColBERT, LTR, ensemble
- •Complete RAG pipeline with document processing and chunking
- •Context retrieval and response generation
- •LLM integration for answer generation with customizable prompt templates
Analytics & Time Series
- •Data analysis: clustering (K-Means, DBSCAN, GMM), dimensionality reduction (PCA)
- •Quality metrics: Recall@K, Precision@K, F1@K, MRR
- •Outlier detection: Z-score, Modified Z-score, IQR
- •Drift detection and topic discovery
- •Time series analysis: ARIMA, forecasting, and seasonal decomposition
Resource Management
- •Comprehensive resources: schema, model catalog, index configs, worker status, system stats
- •Real-time resource subscriptions for live updates
- •Resource discovery and metadata access
- •Search and filtering capabilities for resources
Middleware & Enterprise
- •Pluggable middleware: validation, logging, timeout, error handling
- •Authentication: JWT, API keys, OAuth2 with rate limiting
- •Caching layer with TTL and connection pooling
- •Enterprise features: Prometheus metrics, webhooks, retry/resilience (circuit breaker)
- •Health checks for database, tools, and resources
Dataset Loading & Processing
- •Load from HuggingFace, URLs (CSV, JSON, Parquet), GitHub, S3, and local files
- •Automatic schema detection and optimized PostgreSQL table creation
- •Auto-embedding generation for text columns
- •Batch loading with progress tracking
- •Support for multiple formats with efficient bulk loading
PostgreSQL Tools
- •Complete PostgreSQL administration: version info, database statistics, connection monitoring
- •Lock inspection and replication status
- •Configuration settings and extension management
- •Query performance analysis and system resource monitoring
Comprehensive Tool Suite
- •100+ tools: 27 PostgreSQL administration + 70+ NeuronDB tools
- •Vector operations, ML training, analytics, RAG, reranking, and database management
- •Vector search with 7+ distance metrics
- •Quantization tools: int8, fp16, binary, uint8, ternary, int4
- •Dataset loading from HuggingFace, URLs, GitHub, S3, and local files with auto-embedding
- •Vector graph operations and vecmap (sparse vector) support
- •Complete PostgreSQL administration from version info to performance tuning
Capabilities
MCP server features
| Capability | Description | Performance | Production Ready |
|---|---|---|---|
| MCP Protocol | JSON-RPC 2.0 over stdio | Low-latency communication | ✓ |
| Vector Tools | Search, embeddings, indexing | Sub-second operations | ✓ |
| ML Tools | Training, prediction, evaluation | GPU-accelerated | ✓ |
| Analytics Tools | Clustering, analysis, metrics | Efficient algorithms | ✓ |
| RAG Tools | Document processing, retrieval | End-to-end optimization | ✓ |
| Reranking Tools | Cross-encoder, LLM, Cohere, ColBERT | Neural reranking | ✓ |
| Dataset Loading | HuggingFace, URLs, GitHub, S3, local | Auto-embedding | ✓ |
| Resources | Schema, models, indexes, stats | Real-time subscriptions | ✓ |
| Middleware & Enterprise | Validation, auth, metrics, webhooks | Production-ready | ✓ |
| PostgreSQL Tools | Admin, stats, monitoring | Complete management | ✓ |
Get Started
Connect Claude Desktop to NeuronDB
Deploy NeuronMCP. Enable MCP-compatible clients to access NeuronDB vector search, ML algorithms, and RAG capabilities.