Unread MCP Server

Unread MCP Server – A Read‑Only Search Interface for Your Unread RSS Database

The Unread MCP server addresses a common pain point for developers and power users who rely on the Unread RSS reader: quickly locating specific articles across thousands of feeds without leaving their AI‑assistant workflow. By exposing a lightweight, read‑only interface over Unread’s SQLite database, the server lets Claude or other MCP clients perform fulltext queries directly against article titles, content, authors, and feed names. This eliminates the need for manual browsing or exporting data, enabling AI assistants to surface relevant content on demand.

At its core, the server offers a keyword‑based fulltext search engine that supports Boolean logic (AND, OR, NOT) and exact phrase matching with quotes. Users can filter results by article status—starred, read, or unread—and restrict searches to individual feeds. Each search returns a concise preview (≈300 characters) and an article ID, which can be fed into the companion get‑article tool to retrieve the complete text and metadata. Additional utilities expose database statistics, list all feeds with article counts, and provide a per‑feed search capability. The read‑only nature of the server guarantees that the underlying Unread data remains untouched, making it safe to integrate into automated pipelines.

Developers can leverage this server in a variety of real‑world scenarios. For example, a research assistant might ask Claude to “find unread articles about quantum computing in the MIT Technology Review feed,” and the assistant can immediately return a curated list with links to full texts. Content creators can use it to discover trending topics within their subscribed feeds, while analysts can extract insights from starred or read articles for reporting purposes. The server’s straightforward command‑line interface and dependency on ubiquitous tools (bash, sqlite3, jq) mean it can be deployed quickly on macOS machines already running Unread.

Integration with AI workflows is seamless: once the server is registered in a client’s MCP configuration, it appears as a set of tools that can be invoked through natural language prompts. The search and retrieval capabilities empower assistants to act as a knowledge navigator, pulling specific information from the user’s personal RSS archive without manual intervention. Because the server operates over a local SQLite database, latency is minimal and privacy concerns are mitigated—no external APIs or cloud services are involved.

In summary, the Unread MCP server turns a static RSS reader into an interactive knowledge base that AI assistants can query in real time. Its focus on keyword search, status filtering, and feed scoping provides developers with precise control over content discovery, while its read‑only design ensures data integrity. For anyone looking to embed personalized article search into an AI workflow, this server offers a lightweight, reliable solution that bridges the gap between local RSS consumption and conversational AI.