News MCP Server

This repository demonstrates an example implementation of a Model Context Protocol (MCP) server that exposes tools for querying financial news data from Refinitiv Data Platform (RDP). The project showcases how to integrate news search capabilities with an agentic system built with LangGraph.

What is MCP?

The Model Context Protocol (MCP) is a standardized way to connect AI assistants with data sources and tools. This implementation creates a news server that LLMs can query to get real-time financial news information, enabling AI applications to provide up-to-date market insights.

Project Structure

├── llm.py                    # LLM Configuration
├── chat_app.py               # FastAPI chat interface
├── mcp-servers/
│   ├── news-server.py        # MCP server implementation
│   └── rdp_auth.py          # RDP authentication utilities
├── evals/
│   └── trajectory_llm_as_judge.py  # Evaluation framework
├── pyproject.toml            # Project dependencies
└── langgraph.json           # LangGraph configuration

Configuration

RDP (Refinitiv Data Platform) Credentials

You need valid Refinitiv Data Platform (RDP) credentials with entitlements for news data:

export RDP_USERNAME="your-username"
export RDP_PASSWORD="your-password"  
export RDP_CLIENT_ID="your-client-id"

LangSmith (Optional - for evaluations)

export LANGSMITH_API_KEY="your-langsmith-api-key"
export LANGSMITH_TRACING="true"

Important: Please check your license agreement regarding the use of News data with LLMs before running this application.

For more information, see: https://www.lseg.com/en/data-analytics/financial-news-service/machine-readable-news

Usage

1. MCP Inspector (Development & Testing)

Use the MCP Inspector to test and explore the news MCP server tools:

uv run mcp dev mcp-servers/news-server.py

This opens a web interface where you can:

• Test news search queries
• Explore available tools and their parameters
• View response formats and sample data

2. LangGraph Studio

Launch the interactive LangGraph development environment:

uv run langgraph dev

This provides a visual interface for:

• Building and testing conversation flows
• Debugging agent behavior
• Monitoring tool usage

3. Custom Chat Interface

Start the FastAPI-based chat application:

uv run chat_app.py

Access the chat interface at http://localhost:8000 to interact with the news-enabled AI assistant.

Sample Queries

Try these example queries in any of the interfaces:

"Find recent news about Apple"
"What's the latest on Tesla?"

Evaluation

Run automated evaluations to assess the system's performance. This requires a LANGSMITH account.

uv run pytest evals/trajectory_llm_as_judge.py --langsmith-output