NeuronMCP
Model Context Protocol server for NeuronDB PostgreSQL extension, implemented in Go
Enables MCP-compatible clients to access NeuronDB vector search, ML algorithms, and RAG capabilities.
Power your agent
पब्लिशर का इंस्टॉल। PlugMCP एंडपॉइंट नहीं।
Host a copy with our keys and logs.
रजिस्ट्री में अभी कोई पैकेज्ड रिलीज़ नहीं है। रिपॉज़िटरी का README देखें.
रिमोट URL या आपकी मशीन पर कमांड? दोनों इंस्टॉल में क्या अंतर है
PlugMCP पर चलाएँ
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
लॉन्चNeuronMCP implements the Model Context Protocol using JSON-RPC 2.0 over stdio. It provides tools and resources for MCP clients to interact with NeuronDB, including vector operations, ML model training, and database schema management.
The server registers all available tools at startup (650+ tools: vector, ML, RAG, PostgreSQL administration, and more). There is no environment variable to limit or filter which tools are registered.
Claude Desktop: Some MCP clients, including Claude Desktop, may impose a limit on how many tools they display or use. If you hit that limit, use a different MCP client (e.g. the included neuron-mcp-client) or run the server in a context where the full tool set is needed. See Setup Guide for Claude Desktop configuration.
Example configuration (use the path to your built binary, e.g. ./bin/neuron-mcp):
{
"mcpServers": {
"neurondb": {
"command": "/path/to/neuron-mcp",
"env": {
"NEURONDB_HOST": "localhost",
"NEURONDB_PORT": "5432",
"NEURONDB_DATABASE": "neurondb",
"NEURONDB_USER": "neurondb",
"NEURONDB_PASSWORD": "your_password"
}
}
}
}
https://www.neurondb.ai/docs/neuronmcp — Tool reference, Claude Desktop setup, and configuration.
For a detailed comparison with other MCP servers, see docs/tool-resource-catalog.md and the MCP protocol spec.
graph TB
subgraph CLIENT["MCP Clients"]
CLAUDE[Claude Desktop]
CUSTOM[Custom MCP Clients]
CLI[CLI Tools]
end
subgraph MCP["NeuronMCP Server"]
PROTOCOL[MCP Protocol Handler<br/>JSON-RPC 2.0]
TOOLS[Tool Registry<br/>650+ Tools]
RESOURCES[Resource Manager<br/>Schema, Models, Indexes]
MIDDLEWARE[Middleware Pipeline<br/>Auth, Logging, Rate Limit]
CACHE[TTL Cache<br/>Idempotency]
end
subgraph CATEGORIES["Tool Categories"]
VEC[Vector Operations<br/>50+ tools]
ML[ML Pipeline<br/>25+ algorithm families]
RAG[RAG Operations<br/>Document processing]
PG[PostgreSQL Tools<br/>100+ DDL/DML/DCL<br/>650+ total tools]
DATASET[Dataset Loading<br/>HuggingFace, S3, GitHub]
end
subgraph DB["NeuronDB PostgreSQL"]
VECTOR[Vector Search<br/>HNSW/IVF]
EMBED[Embeddings<br/>Text/Image/Multimodal]
ML_FUNC[ML Functions<br/>25+ algorithm families]
ADMIN[PostgreSQL Admin<br/>Full DDL/DML/DCL]
end
CLAUDE -->|stdio| PROTOCOL
CUSTOM -->|stdio/HTTP/SSE| PROTOCOL
CLI -->|stdio| PROTOCOL
PROTOCOL --> MIDDLEWARE
MIDDLEWARE --> CACHE
CACHE --> TOOLS
CACHE --> RESOURCES
TOOLS --> VEC
TOOLS --> ML
TOOLS --> RAG
TOOLS --> PG
TOOLS --> DATASET
VEC --> VECTOR
ML --> ML_FUNC
RAG --> EMBED
PG --> ADMIN
DATASET --> VECTOR
style CLIENT fill:#e3f2fd
style MCP fill:#fff3e0
style CATEGORIES fill:#f3e5f5
style DB fill:#e8f5e9
sequenceDiagram
participant Client as MCP Client
participant Server as NeuronMCP Server
participant Tools as Tool Registry
participant DB as NeuronDB
Client->>Server: Initialize (JSON-RPC)
Server-->>Client: Server Capabilities
Client->>Server: tools/list
Server->>Tools: Get available tools
Tools-->>Server: Tool catalog
Server-->>Client: Tool list (650+ tools)
Client->>Server: tools/call {"name": "vector_search", ...}
Server->>Server: Validate & authenticate
Server->>Tools: Execute tool
Tools->>DB: Execute SQL query
DB-->>Tools: Query results
Tools-->>Server: Tool response
Server-->>Client: JSON-RPC response
Note over Client,DB: Streaming supported via SSE
graph LR
subgraph TOOLS["650+ Tools"]
VEC_TOOLS[Vector Operations<br/>Search, Embeddings, Indexing]
ML_TOOLS[ML Pipeline<br/>25+ algorithm families<br/>Training, Prediction, Evaluation]
RAG_TOOLS[RAG Operations<br/>Document Processing<br/>Context Retrieval]
PG_TOOLS[PostgreSQL Tools<br/>100+ tools<br/>DDL, DML, DCL, Admin]
DATASET_TOOLS[Dataset Loading<br/>HuggingFace, S3, GitHub<br/>Auto-embedding]
end
style VEC_TOOLS fill:#ffebee
style ML_TOOLS fill:#e8f5e9
style RAG_TOOLS fill:#fff3e0
style PG_TOOLS fill:#e3f2fd
style DATASET_TOOLS fill:#f3e5f5
Use the Setup Guide for Claude Desktop. If your client limits the number of tools, use another MCP client such as the included
neuron-mcp-client.
Option 1: Using Docker Compose (recommended for quick start)
If you have PostgreSQL with NeuronDB (e.g. from the neurondb repo Docker setup):
# Create extension if not already created
psql "postgresql://neurondb:neurondb@localhost:5433/neurondb" -c "CREATE EXTENSION IF NOT EXISTS neurondb;"
Option 2: Native PostgreSQL Installation
createdb neurondb
psql -d neurondb -c "CREATE EXTENSION neurondb;"
NeuronMCP can use an optional database schema for LLM model config, API keys, index templates, worker settings, and tool defaults. Run ./scripts/neuronmcp-setup.sh from the repository root to set it up. See neurondb-mcp-setup.md for details.
Create mcp-config.json:
{
"database": {
"host": "localhost",
"port": 5433,
"database": "neurondb",
"user": "neurondb",
"password": "neurondb"
},
"server": {
"name": "neurondb-mcp-server",
"version": "2.0.0"
},
"logging": {
"level": "info",
"format": "text"
},
"features": {
"vector": { "enabled": true },
"ml": { "enabled": true },
"analytics": { "enabled": true }
}
}
Or use environment variables:
export NEURONDB_HOST=localhost
export NEURONDB_PORT=5432
export NEURONDB_DATABASE=neurondb
export NEURONDB_USER=neurondb
export NEURONDB_PASSWORD=neurondb
From the repository root:
make build
export NEURONDB_HOST=localhost NEURONDB_PORT=5432 NEURONDB_DATABASE=neurondb NEURONDB_USER=neurondb NEURONDB_PASSWORD=neurondb
./bin/neuron-mcp
Optional: pass a config file with -c or set NEURONDB_MCP_CONFIG to the path of your mcp-config.json.
Test: run the included client to list tools:
./bin/neuron-mcp-client ./bin/neuron-mcp tools/list
Or send a raw JSON-RPC initialize to confirm the server responds:
echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}' | ./bin/neuron-mcp
Use the setup script for database schema (optional LLM/config tables):
# From repository root
./scripts/neuronmcp-setup.sh
# With system service enabled (if supported)
./scripts/neuronmcp-setup.sh --enable-service
Use scripts/neuronmcp-run.sh or scripts/neuronmcp-run-server.sh to run the server.
go build -o bin/neuron-mcp ./cmd/neurondb-mcp
./bin/neuron-mcp
This repo does not include a docker-compose.yml. Build and run the image manually:
# From repository root
docker build -f docker/Dockerfile -t neurondb-mcp:latest .
docker run -i --rm \
-e NEURONDB_HOST=localhost \
-e NEURONDB_PORT=5432 \
-e NEURONDB_DATABASE=neurondb \
-e NEURONDB_USER=neurondb \
-e NEURONDB_PASSWORD=neurondb \
neurondb-mcp:latest
For full-stack Docker (NeuronDB + NeuronMCP), use the neurondb repository or deploy each component from its repo.
For systemd (Linux) or launchd (macOS), see your system documentation or the neurondb installation services guide for patterns.
NeuronMCP uses Model Context Protocol over stdio:
Example request:
{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "vector_search",
"arguments": {
"query_vector": [0.1, 0.2, 0.3],
"table": "documents",
"limit": 10
}
}
}
See mcp-config.json.example for complete configuration structure. Environment variables override configuration file values.
NeuronMCP provides comprehensive tools covering all NeuronDB capabilities:
Tool reference: Tool & Resource Catalog. PostgreSQL tools: docs/postgresql-tools.md.
For example client usage and interaction transcripts, see docs/examples/.
The load_dataset tool supports multiple data sources with automatic schema detection, embedding generation, and index creation:
{
"name": "load_dataset",
"arguments": {
"source_type": "huggingface",
"source_path": "sentence-transformers/embedding-training-data",
"split": "train",
"limit": 10000,
"auto_embed": true,
"embedding_model": "default"
}
}
{
"name": "load_dataset",
"arguments": {
"source_type": "url",
"source_path": "https://example.com/data.csv",
"format": "csv",
"auto_embed": true,
"create_indexes": true
}
}
{
"name": "load_dataset",
"arguments": {
"source_type": "github",
"source_path": "owner/repo/path/to/data.json",
"auto_embed": true
}
}
{
"name": "load_dataset",
"arguments": {
"source_type": "s3",
"source_path": "s3://my-bucket/data.parquet",
"auto_embed": true
}
}
{
"name": "load_dataset",
"arguments": {
"source_type": "local",
"source_path": "/path/to/local/file.jsonl",
"schema_name": "my_schema",
"table_name": "my_table",
"auto_embed": true
}
}
Key Features:
NeuronMCP exposes the following resources:
NeuronMCP is fully compatible with Claude Desktop on macOS, Windows, and Linux.
Create Claude Desktop configuration file:
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
Linux: ~/.config/Claude/claude_desktop_config.json
See the example configuration files in this directory (claude_desktop_config.*.json) for platform-specific examples.
Example configuration:
{
"mcpServers": {
"neurondb": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"--network", "neurondb-network",
"-e", "NEURONDB_HOST=neurondb-cpu",
"-e", "NEURONDB_PORT=5432",
"-e", "NEURONDB_DATABASE=neurondb",
"-e", "NEURONDB_USER=neurondb",
"-e", "NEURONDB_PASSWORD=neurondb",
"neurondb-mcp:latest"
]
}
}
}
Or use local binary:
{
"mcpServers": {
"neurondb": {
"command": "/path/to/neuron-mcp",
"env": {
"NEURONDB_HOST": "localhost",
"NEURONDB_PORT": "5432",
"NEURONDB_DATABASE": "neurondb",
"NEURONDB_USER": "neurondb",
"NEURONDB_PASSWORD": "neurondb"
}
}
}
}
Use the full path to your built binary (e.g. /home/user/neuron-mcp/bin/neuron-mcp).
Restart Claude Desktop after configuration changes.
Run NeuronMCP interactively for testing:
./bin/neuron-mcp
Send JSON-RPC messages via stdin, receive responses via stdout.
A simple MCP client that works like Claude Desktop. It handles the full MCP protocol including the initialize handshake.
Build the client (included in make build):
make build
Usage:
# Initialize and list tools
./bin/neuron-mcp-client ./bin/neuron-mcp tools/list
# Call a tool
./bin/neuron-mcp-client ./bin/neuron-mcp tools/call '{"name":"vector_search","arguments":{}}'
# List resources
./bin/neuron-mcp-client ./bin/neuron-mcp resources/list
The client automatically:
Test script:
cd src/client
./example_usage.sh
Or use the Python client:
cd src/client
python neurondb_mcp_client.py -c ../tests/neuronmcp_server.json -e "list_tools"
For Docker:
docker run -i --rm \
-e NEURONDB_HOST=localhost \
-e NEURONDB_PORT=5432 \
-e NEURONDB_DATABASE=neurondb \
-e NEURONDB_USER=neurondb \
-e NEURONDB_PASSWORD=neurondb \
neurondb-mcp:latest
Build the image first: docker build -f docker/Dockerfile -t neurondb-mcp:latest . from the repository root.
Related: NeuronDB (extension), NeuronAgent (agent runtime).
NeuronMCP requires PostgreSQL 16+ with the NeuronDB extension. Install NeuronDB first: neurondb repository (Simple Start). Full-stack deployment (NeuronDB + NeuronMCP + other components) is documented in each component’s repository.
Ensure stdin and stdout are not redirected:
./bin/neuron-mcp # Correct
./bin/neuron-mcp > output.log # Incorrect - breaks MCP protocol
For Docker, use interactive mode:
docker run -i --rm neurondb-mcp:latest
Verify connection parameters:
psql -h localhost -p 5432 -U neurondb -d neurondb -c "SELECT 1;"
Check environment variables:
env | grep NEURONDB
Verify container is running (if using Docker from another repo):
docker ps | grep neurondb-mcp
Test stdio manually:
echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}' | ./bin/neuron-mcp
Check client configuration file path and format.
Verify config file path:
ls -la mcp-config.json
Check environment variable names (must start with NEURONDB_):
env | grep -E "^NEURONDB_"
See LICENSE for license information.
MCP server for PostgreSQL and NeuronDB (PostgreSQL AI System)
टूल
अभी टूल सूची नहीं है। रिमोट सर्वर को पहले हेल्थ चेक के बाद सूची मिलती है।
FAQ
इस पेज पर Cursor, Claude Desktop, Claude Code या VS Code के लिए stdio कॉन्फ़िग कॉपी करें। यह पब्लिशर का इंस्टॉल इस्तेमाल करता है, PlugMCP प्रॉक्सी नहीं।
इस रिकॉर्ड में कोई ज़रूरी एनवायरनमेंट वेरिएबल दर्ज नहीं है।
Neuron stdio पर लोकल चलता है।
पिछली जाँच 2026-10-28 को हुई। MCP server for PostgreSQL and NeuronDB (PostgreSQL AI System)