It’s been about a year since we rolled out GeoFeeds, a spatial new aggregator along the lines of the old Planet Geospatial. During that time, it’s been humming along, and we’ve added about 90 blog feeds to it. It provides a single, rolling, aggregated feed of posts from those blogs over the previous year. I’ve been impressed by the surge is RSS feeds over that time.
If you have a blog or feed related to geospatial, personal or corporate, you can simply add an issue to the repo, including the title, and feed URL, and it will get added to the OPML.
Today, I pushed an MCP endpoint to production. This enables your AI client, such as Claude Desktop, to analyze the cached feed, providing an additional way to monitor the state of the geospatial market as it evolves in real time. For example, I asked Claude to tell me the most prolific blogs so far this month (January, 2026). I spot checked a few of these and the numbers were right. (It also correctly flagged OSGeo’s spam problem.)
It also went a little further and provided some interesting highlights from the less prolific blogs.
The MCP server is deployed using streamable HTTP, which should make it fairly easy to configure. In Claude Desktop, you do the following:
- Go to Settings → Connectors
- Scroll to “Add custom connector”
- Enter the URL:
https://geofeeds.me/mcp - Click Add
That’s it. No need for editing JSON config files for the desktop app. There’s no authentication since this is a public, read-only resource. It’s deployed using the serverless App Platform from DigitalOcean, so it should be fairly robust. Here is the list of the tools the server exposes. I have gotten the best result working with the aggregated feed and the feed items. The cached feeds can have some repetition that can throw off a loosely constructed prompt.
I’ve been meaning to get to doing this for a few months, but it kept falling off the plate. I have recently had a need to dig into to doing MCP with Node, instead of Python, so this was a good first cut. I’m happy with the result.
