Version:

Building from Source

Deployment of the pgEdge Postgres MCP Server is easy; you can get up and running in a test environment in minutes.

Note

The MCP server is also available as a supporting component for pgEdge Enterprise Postgres. For details, see the platform-specific documentation for Postgres Enterprise Postgres.

Before deploying the server from source, you need to install and obtain:

a Postgres database (with pg_description support)
an LLM Provider API key: Anthropic, OpenAI, or Ollama (local/free)

In your Postgres database, you'll create a LOGIN user; the user name and password will be shared in the configuration file used for deployment.

After meeting the prerequisites, use the steps that follow to build the MCP server.

Clone the Repository

To build from source, first, clone the pgedge-postgres-mcp repository and navigate into the repository's root directory:

git clone https://github.com/pgEdge/pgedge-postgres-mcp.git
cd pgedge-postgres-mcp

Then, build the pgedge-postgres-mcp binary; the file is created in the bin directory under your current directory:

make build

Create a Configuration File

The .env.example file contains a sample configuration file that we can use for deployment; instead of updating the original, we copy the sample file to bin/pgedge-postgres-mcp.yaml:

cp .env.example bin/pgedge-postgres-mcp.yaml

Then, edit the configuration file, adding deployment details. In the DATABASE CONNECTION section, provide Postgres connection details:

# ============================================================================
# DATABASE CONNECTION
# ============================================================================
# PostgreSQL connection details
PGEDGE_DB_HOST=your-postgres-host
PGEDGE_DB_PORT=5432
PGEDGE_DB_NAME=your-database-name
PGEDGE_DB_USER=your-database-user
PGEDGE_DB_PASSWORD=your-database-password
PGEDGE_DB_SSLMODE=prefer

Specify the name of your embedding provider in the EMBEDDING PROVIDER CONFIGURATION section:

# ============================================================================
# EMBEDDING PROVIDER CONFIGURATION
# ============================================================================
# Provider for text embeddings: voyage, openai, or ollama
PGEDGE_EMBEDDING_PROVIDER=voyage

# Model to use for embeddings
# Voyage: voyage-3, voyage-3-large (requires API key)
# OpenAI: text-embedding-3-small, text-embedding-3-large
#         (requires API key)
# Ollama: nomic-embed-text, mxbai-embed-large (requires local Ollama)
PGEDGE_EMBEDDING_MODEL=voyage-3

Provide your API key in the LLM API KEYS section:

# ============================================================================
# LLM API KEYS
# ============================================================================
# Anthropic API key (for Claude models and Voyage embeddings)
# Get your key from: https://console.anthropic.com/
PGEDGE_ANTHROPIC_API_KEY=your-anthropic-api-key-here

# OpenAI API key (for GPT models and OpenAI embeddings)
# Get your key from: https://platform.openai.com/
PGEDGE_OPENAI_API_KEY=your-openai-api-key-here

# Ollama server URL (for local models)
# Default: http://localhost:11434 (change if Ollama runs elsewhere)
PGEDGE_OLLAMA_URL=http://localhost:11434

During deployment, users are created for the deployment; you can specify user information in the AUTHENTICATION CONFIGURATION section. For a simple test environment, the INIT_USERS property is the simplest configuration:

# ============================================================================
# AUTHENTICATION CONFIGURATION
# ============================================================================
# The server supports both token-based and user-based authentication
# simultaneously. You can initialize both types during container
# startup.

# Initialize tokens (comma-separated list)
# Use for service-to-service authentication or API access
# Format: token1,token2,token3
# Example: INIT_TOKENS=my-secret-token-1,my-secret-token-2
INIT_TOKENS=

# Initialize users (comma-separated list of username:password
# pairs)
# Use for interactive user authentication with session tokens
# Format: username1:password1,username2:password2
# Example: INIT_USERS=alice:secret123,bob:secret456
INIT_USERS=

# Client token for CLI access (if using token authentication)
# This should match one of the tokens in INIT_TOKENS
MCP_CLIENT_TOKEN=

You also need to specify the LLM provider information in the LLM CONFIGURATION FOR CLIENTS section:

# ============================================================================
# LLM CONFIGURATION FOR CLIENTS
# ============================================================================
# Default LLM provider for chat clients: anthropic, openai, or
# ollama
PGEDGE_LLM_PROVIDER=anthropic

# Default LLM model for chat clients
# Anthropic: claude-sonnet-4-20250514, claude-opus-4-20250514,
#             etc.
# OpenAI: gpt-5-main, gpt-4o, gpt-4-turbo, etc.
# Ollama: llama3, mistral, etc.
PGEDGE_LLM_MODEL=claude-sonnet-4-20250514

Use the Command Line to Create a User

# Add a user for web access
./bin/pgedge-postgres-mcp -add-user admin -user-password \
"your_password"

Deploy the Server

Then, use the following command to start the pgedge-postgres-mcp:

./bin/pgedge-postgres-mcp

Connect with a Browser and Authenticate

Then, use your browser to open http://localhost:8080 and authenticate with the server.

You're ready!

Start asking questions about your database in natural language.