reltio-ai/reltio-mcp-server

Reltio MCP Server – Developer Edition

Reltio MCP Server is a lightweight, plugin-based Model Context Protocol (MCP) server designed to perform advanced entity matching with language models in Reltio environments.

DISCLAIMER: By entering in your credentials and consuming the Reltio MCP Server – Developer Edition, you may be connecting to the Reltio Platform via API calls. Any and all API calls will be counted against the API call entitlements set out in your SaaS Subscription Agreement (the "Agreement") with Reltio. API calls in excess of those entitlements, including any API calls made while consuming the server, could result in Overages, as set out in the Agreement.

Table of Contents

• Available Tools
• Environment Configuration
• Server Prerequisites
• Client Prerequisites
• Running the Server
• Integration with Claude Desktop App
• Custom MCP Client Integration
• Testing
• Agent Client

Available Tools

Environment Configuration

Create a .env file in the root directory:

RELTIO_SERVER_NAME=RELTIO_MCP_SVR_NAME
RELTIO_ENVIRONMENT=RELTIO_ENVIRONMENT
RELTIO_CLIENT_ID=RELTIO_CLIENT_ID
RELTIO_CLIENT_SECRET=RELTIO_CLIENT_SECRET
RELTIO_TENANT=RELTIO_TENANT
RELTIO_AUTH_SERVER=RELTIO_AUTH_SEVER # Default: https://auth.reltio.com

Server Prerequisites

• Python ≥ 3.10
• (Optional) Docker installed, only if you choose to run the server with Docker - See instructions below
• (Optional) uv (python package manager) installed, only if you choose to run the server without Docker - See instructions below

Client Prerequisites

If you want to consume the server from Claude AI (Anthropic):

• Claude Desktop App installed from claude.ai
• Node.js

If you want to consume the server from a custom client (OpenAI, Gemini, Anthropic)

• Go to the section below on Custom MCP Client Integration

Running the Server

Option 1: Manual Execution (RECOMMENDED)

Run this script to automate setup (virtualenv, install, Claude config injection):

If you are on Windows, open a terminal and run this command:

⚠️ Right-click setup.bat → Properties → Unblock → Run as Admin (if SmartScreen warning appears)

setup.bat

If you are on macOS / Linux, open a terminal and run this command:

bash setup.sh

You then might need to activate your newly-created virtual environment if it was not activated in your terminal already

source .venv/bin/activate

After initial setup (it applies to Windows/macOS/Linux), run this command:

mcp install --with requests --with pyyaml main.py -f .env 

Open Claude (or your custom MCP Client) and start using it.

Option 2: With Docker

Make sure you have Docker installed (https://www.docker.com/products/docker-desktop/)

Open a terminal and run this command:

docker compose up -d --build

To configure Claude to use this MCP server, please go into the next section.

Integration with Claude Desktop App

You must follow these steps ONLY if you are running the server with docker (option 2 above). Option 1 does not require this subsequent step.

Step 1: Launch Claude Desktop

Open the Claude app on your system.

Step 2: Access Settings

If you are on Windows

1. Open Claude Desktop.
2. Click the hamburger menu (top-left).
3. Go to: Settings → Developer → Edit Config.

If you are on macOS

1. Open Claude Desktop.
2. Click Claude from the macOS menu bar.
3. Go to: Settings → Developer → Edit Config.

Add this entry to the claude_desktop_config.json file:

{
  "mcpServers": {
    "reltio-mcp-server": {
      "command": "npx",
      "args": [
        "mcp-remote@0.0.22",
        "http://localhost:8000/sse"
      ]
    }
  }
}

Always Restart the Claude Desktop app after making changes.

NOTE: ECONNREFUSED in Claude Logs? Check mcp-remote

If Claude logs show ECONNREFUSED, the mcp-remote service (possibly at http://localhost:8000/sse) might be the issue.

🔪 Kill mcp-remote Process

🐧 Linux / macOS

ps aux | grep "mcp-remote http://localhost:8000/sse" | grep -v grep | awk '{print $2}' | xargs kill

🪟 Windows (PowerShell)

Get-CimInstance -ClassName Win32_Process -Filter "CommandLine LIKE '%mcp-remote http://localhost:8000/sse%'" | ForEach-Object { Stop-Process -Id $_.ProcessId -Force }

Custom MCP Client Integration

Install dependencies:

pip install -r requirements.txt

Ensure your MCP server is running locally before executing clients.

Claude MCP Client

python ./clients/mcp_claude_client.py

You will need:

• Anthropic API Key (sk-ant-api...)
• Claude Model ID (e.g., claude-3-opus-20240229)
• MCP Server URL (e.g., http://localhost:8000/sse)

Gemini MCP Client

python ./clients/mcp_gemini_client.py

Required:

• Google API Key (AIza...)
• Gemini Model ID (e.g., gemini-1.5-pro)
• MCP Server URL

OpenAI MCP Client

python ./clients/mcp_openai_client.py

Required:

• OpenAI API Key (sk-...)
• OpenAI Model ID (gpt-4-turbo)
• MCP Server URL

Testing

Run Tests via Docker

docker-compose -f docker-compose-test.yaml up -d --build

Or Run Locally

pip install -r requirements_tests.txt
./run_tests.sh --coverage

Test Suite

• tests/unit/test_server.py
• tests/unit/test_server_structure.py
• tests/unit/test_server_error_handling.py
• tests/unit/test_tools_activity.py
• tests/unit/test_main.py

Agent Client

A CLI chat bot that connects to Reltio MCP server using OAuth 2.0 authentication and provides an interactive interface for querying Reltio data.

Prerequisites

cd clients/agent_with_mcp
pip install -r requirements.txt

Configuration

Edit the configuration constants in agent_client.py:

# Reltio Configuration
NAMESPACE = "your_namespace"  # Your Reltio namespace
RELTIO_CLIENT_ID = "your_client_id"  # Your Reltio client ID
RELTIO_CLIENT_SECRET = "your_client_secret"  # Your Reltio client secret

# Model Configuration
MODEL_ID = "anthropic:claude-3-5-sonnet-20241022"  # Supported: anthropic, google_genai, openai
API_KEY = "your_api_key"  # API key for the model provider

Running the Agent Client

python agent_client.py

The client will:

1. Open a browser for OAuth authentication
2. Establish connection to Reltio MCP server
3. Provide an interactive chat interface

Supported Models

• Anthropic: anthropic:claude-3-5-sonnet-20241022
• Google: google_genai:gemini-2.0-flash-001
• OpenAI: openai:gpt-4o-mini