Hi friends, I built a tiny macOS app called Warplet (https://warplet.app/) that connects remote LLMs like Claude, ChatGPT, Grok or Perplexity to local MCP servers like the one from DEVONthink.
(If this kind of post is against the forum rules please let me know and I’ll take it down. I wanted to share it here anyway because I specifically built it for my own DEVONthink workflow)
How it works:
- You setup Warplet and tell it that DEVONthink MCP server is running for example on
http://localhost:8420
- Warplet gives you a remote URL that you plug into Claude, Grok, Perplexity, ChatGPT and so on, like in my case
https://serene-wind-5hxm.warplet.app/devonthink - On connect, Warplet handles OAuth and login and completes the OAuth2 handshake with the remote LLM, and ensures request conform the request shape to all the popular providers (especially Perplexity was a huge pain to get supported properly…)
- When the AI tries to connect to the MCP server, Warplet forwards all the requests directly to the DEVONthink MCP that’s running on your Mac, which then handles your query
Why though?
This allows you to have Warplet + DEVONthink running on a unused computer or server and remote control it from any of the AI web versions or apps. It’s effectively a way to get Claude Cowork + Dispatch functionality (where you have a computer running Claude), but in all of the LLMs and not just in Claude.
Currently the only Desktop grade app that can access local MCP servers is Claude Desktop. Perplexity used to be able to do that too, but since they rewrote their macOS app, they removed local MCP functionality completely.
These days I use Grok as my main AI, and as it doesn’t have local MCP access, it left a notable gap in my workflow.
Some examples
Here’s me using Grok on the web to classify my inbox:
Here I use Perplexity on the web with “Deep Research” to do a broad research across my entire DEVONthink database to aggregate my ad spending. This ran for around 3 minutes in the background while I was doing something else
Here I used the ChatGPT iOS app to ask about my boarding time for a previous flight. ChatGPT uses DEVONthink through Warplet to find my boarding pass, read it, and give me the details
How much?
The app uses a remote server to forward requests so I sadly can’t provide it for free (yet, though I hope to have a limited free version up soon as well). It’s currently $1.99/mo + 1 week free trial to test the app.
Inspiration
A lot of inspiration to build this came from [Release] DEVONthink MCP Server - #7 by syntagm which had a lot of cool workflows people came up with for this MCP, and direct messages I received personally to thank me for building it.
Notable shoutout to @rkaplan for the idea to use Dispatch to access DEVONthink on the go, and his Perplexity Max research experiments
Feedback wanted
The app is brand new, I literally just got it approved by Apple. I am sure there are things that don’t work yet but I did my best to test it with any AI I have access to that allows adding of remote MCP connectors.
I am also not sure about the price. I want to offer a free version but need to first see how much monthly server cost this is actually incurring before I can do that.
Any feedback is greatly appreciated!
