hafnium49/mcp_server_feedly

MCP Server for Feedly API

This repository contains a minimal typescript-sdk server that exposes selected Feedly API endpoints as MCP tools. The server provides the following tools:

• feedly.search
• feedly.collect
• feedly.entity_lookup
• feedly.autocomplete

These tools make it possible for MCP-aware language models to search and retrieve articles or NLP entity information from Feedly.

Complete Feedly Workflow

The tools work together in a typical discovery loop:

1. Entity discovery – use feedly.autocomplete to find IDs for relevant topics or companies.
2. Content search – call feedly.search with those IDs to locate articles and discover stream IDs.
3. Stream collection – pass a stream ID to feedly.collect to retrieve full articles from a feed or board.
4. Entity details – fetch additional information about any entity using feedly.entity_lookup.

Key Concepts

• Entities – topics, companies, people or technologies identified by Feedly IDs.
• Streams – RSS feeds, user categories or publication buckets.
• Salience – choose mention or about to control how prominently an entity appears in results.
• Sources – which streams to search in (defaults to all topics).

Common entity ID patterns include nlp/f/topic/xxxx for topics and feed/http://... for RSS feeds.

Troubleshooting

• Ensure FEEDLY_TOKEN is exported so the server can call the Feedly API.
• Use feedly.autocomplete first to confirm entity IDs.
• Some streams may not support direct collection; discover stream IDs via feedly.search.
• Use pagination tokens and the count parameter to stay within Feedly rate limits.

Example Workflows

The comments in server.ts contain full examples such as tracking AI in healthcare, running company research and analyzing technology trends.

Setup

1. Install dependencies using npm:

npm install

2. Export your Feedly authentication token so the server can call the Feedly API:

export FEEDLY_TOKEN=YOUR_TOKEN_HERE

Running the server

Run the server over stdio so there is no network port to configure. Simply start it with:

npx ts-node server.ts

The process stays alive and communicates with clients via its standard input and output streams.

Using with Claude Desktop

To connect Claude Desktop directly to this server, open File > Settings > Developer > Edit Config and create or update claude_desktop_config.json with an entry like:

{
  "mcpServers": {
    "feedly": {
      "command": "npx",
      "args": [
        "ts-node",
        "/path/to/mcp_server_feedly/server.ts"
      ],
      "env": {
        "FEEDLY_TOKEN": "YOUR_TOKEN_HERE"
      }
    }
  }
}

Claude Desktop communicates with the server over stdio, so there is no URL to configure. command and args tell the app how to start it. Replace the path with the location of server.ts on your system and set FEEDLY_TOKEN to your Feedly token.

You can copy claude_desktop_config.example.json from this repository as a starting point.

Running tests

Install the Python dependencies and run the test suite:

pip install -r requirements.txt
FEEDLY_TOKEN=xxxx pytest -q