Developer builds AI news aggregator with MCP, DeepSeek and a Telegram bot

This project didn't start as an abstract idea. It started with a concrete problem: on February 28, 2026, the author was in Dubai while Iranian drones were being intercepted, and he couldn't quickly get a coherent picture of what was happening. The result is a multilingual news aggregator that collects news, translates it, tags it and serves it not only to people but also to AI assistants.

How the project started

The author describes the starting point in very down-to-earth terms: Google and Yandex surfaced stories that were two hours old, while Telegram channels offered contradictory accounts of events. When you need news right now, that kind of lag stops being a minor inconvenience and becomes a real problem. The first fix was simple: a Middle East Telegram channel fed manually with RSS streams from major global news agencies.

The manual approach quickly hit a volume ceiling. The stream grew to 200+ posts a day, and keeping it going without automation no longer made sense. So came a parser, then automatic translation, and then IT sources and economic coverage. What began as a local idea for a breaking-news channel turned into a full product: an aggregator site with a news feed, AI chat, an API, an MCP server and a separate Telegram bot.

What's inside the system

The project is built as a set of independent components rather than a single monolith. A dedicated FeedManager service pulls RSS, processes the feeds and writes the data to MySQL. The Flask side reads finished articles and doesn't care whether the parser has stalled on some flaky source. Automatic tagging runs as its own process: it reads fresh stories from the database and writes the results back, while the Telegram bot answers users in its own loop.

• 80+ sources in a single stream, including international agencies, IT media and business publications
• 9 topic categories and about 50 tags for navigating the feed
• automatic translation of stories into Russian immediately after publication
• a REST API with search and filters by tag, date and source
• an MCP server that lets AI assistants use the news database as a tool

The API and MCP server sit on top of the aggregator and serve external clients and AI agents. The key detail is that they don't just return raw JSON: they deliver structured responses through Pydantic models such as ArticleSummary and SearchResult. That cuts down on processing errors and saves tokens. As the author designed it, an assistant can find the right tag on its own, request articles from the past few days, read the full texts and assemble a digest without any manual data prep. Tools are picked up dynamically every five minutes, so new capabilities can be added without rewriting the agent's logic.

Why there's no unnecessary AI

The project doesn't use AI everywhere, only where it actually pays off. The chat interface runs on DeepSeek: according to the author, it's 15–30 times cheaper than the alternatives with comparable quality for news search and summarization. For a free product, that's critical. The same pragmatism shows up in how translations are stored: the multilingual versions of each article live directly in a JSON field on the record, which speeds up access and makes it easier to scale to new languages.

The most telling trade-off is automatic tagging. Instead of an LLM, the author used keyword matching across 47 topics and roughly 600 keywords in English and Russian. This approach achieves about 85% accuracy, takes less than a millisecond per article and spends no tokens at all. Against the widespread urge to "stuff a neural network into every step," it looks like a mature engineering stance: AI belongs where it adds value, not where it merely looks trendy.

What it means

This case shows that a news AI product today isn't just a model that can summarize text. The real value comes from reliable data collection, sound structure, a clear API and a protocol like MCP for plugging in assistants. For media outlets, research teams and internal analytics services, the takeaway is clear: an ordinary news feed can become working infrastructure for live digests, search and automated answers.