news-agent-mcp-server

🧠 News Agent MCP Server

1. Introduction

The News Agent MCP Server is a lightweight, schema-backed capability engine designed for LLM-based agents. It provides structured and semi-structured information retrieval tools over a streamable HTTP interface, enabling agents to perform iterative research workflows.

This server is not intended to be exposed as a public API. Instead, it operates as an internal execution layer within a larger system, where orchestration, safety, and control are handled externally.

---

2. System Role in Architecture

This MCP server is part of a modular LLM system with clear separation of responsibilities:

• Agent (LLM Layer)
  Handles reasoning, decision-making, and tool selection

• MCP Server (This Repository)
  Executes capabilities such as news retrieval and web scraping

• Gateway Layer (Go-based)
  Manages rate limiting, authentication, routing, and safety constraints

This separation ensures:

• Cleaner system design
• Easier scalability
• Independent evolution of components

---

3. Design Philosophy

The server is intentionally designed with the following principles:

• Minimal Core Engine
  Focus only on capability execution, not orchestration or control

• Schema Where Needed, Flexibility Where Needed
  • Structured outputs for predictable data (news)
  • Flexible outputs for rich content (web scraping)

• LLM-First Interface
  All inputs and outputs are strings to ensure seamless integration with LLM agents

• Externalized Concerns
  Features like rate limiting, authentication, and recursion control are handled outside this service

---

4. Architecture Overview

• Built using FastMCP (Python)
• Tools are exposed via @tool decorators
• Runs as a streamable HTTP server
• Endpoint exposed at: /mcp
• Designed to be deployed as a Docker container

---

5. Capabilities

5.1 get_latest_news — Discovery Layer

Fetches recent news articles based on a given topic.

Input

• topic (string)

Output

• JSON string (serialized)

Internally backed by a Pydantic model:

class Article(BaseModel):
    article_id: Optional[str] = None
    link: Optional[str] = None
    title: Optional[str] = None
    description: Optional[str] = None
    creator: Optional[list[str]] = None
    country: Optional[list[str]] = None
    category: Optional[list[str]] = None
    pubDate: Optional[str] = None
    image_url: Optional[str] = None
    source_url: Optional[str] = None

External Dependency:

• newsdata.io

Role:

• Acts as the entry point for discovery
• Provides structured, machine-consumable data for agents

---

## 6. Deep Research Pattern

This server is designed to support iterative research workflows:

Topic → News → URLs → Scrape → Extract Insights → Discover New URLs → Repeat (bounded)

This pattern enables agents to:

• Start with high-level discovery
• Expand into detailed content
• Traverse information progressively

The MCP server provides the building blocks, while:

• The agent controls reasoning
• The gateway enforces limits and safety

---

7. Interface Contract

Input/Output Design

• All tools accept string inputs
• All tools return string outputs

Internal vs External Representation

Layer	Representation
Internal	Structured (Pydantic models)
External	Serialized strings

Rationale

• Eliminates parsing friction for LLMs
• Keeps integration simple and predictable
• Maintains flexibility across tools

---

8. Setup & Deployment

Environment Variables

NEWSDATA_API_KEY=your_api_key
FIRECRAWL_API_KEY=your_api_key

---

Running with Docker

The server is containerized and can be run using Docker:

docker run -p <host_port>:<container_port> <image_name>

Endpoint

/mcp

---

9. Limitations (Intentional)

This server intentionally avoids implementing:

• Rate limiting
• Authentication
• Request validation layers
• Recursion or depth control

Why?

These concerns are handled by the upstream gateway layer, allowing this service to remain:

• Simple
• Focused
• Easy to maintain

---

10. Future Enhancements

Potential improvements include:

• Optional structured output mode for all tools
• Tool metadata and schema exposure for dynamic discovery
• Built-in safety primitives (optional, toggle-based)
• Observability hooks (logging, tracing)
• Configurable content limits (size, depth, etc.)

---

Final Note

This MCP server is designed as a modular capability engine, not a full application. Its strength lies in:

• Clean separation of concerns
• LLM-friendly interface design
• Support for iterative, multi-step research workflows

It is best used as part of a larger system where reasoning, control, and safety are handled externally.

---

👤 Maintainer

Tuhin Kumar Dutta

• 🌐 Website: https://www.tuhindutta.com/
• 💼 LinkedIn: https://www.linkedin.com/in/t-k-dutta

---

⭐ Contribute

Pull requests and issues are welcome.

git clone https://github.com/tuhindutta/news-agent-mcp-server.git

DockerHub: https://hub.docker.com/r/tkdutta/news-agent-mcp-server

This site is open source. Improve this page.