Great job on the MCP!
I’m still testing it out, but I appreciate that you decided to do it.
I’ve really gotten into Hermes Agent, and working through MCP will help a lot with organizing the flood of information that I currently have to deal with..
Great job on the MCP!
I’m still testing it out, but I appreciate that you decided to do it.
I’ve really gotten into Hermes Agent, and working through MCP will help a lot with organizing the flood of information that I currently have to deal with..
Thank you for the kind words, really appreciated.
Thanks +1. It’s nuts that I can use an external AI to talk to DEVONthink, where it will use DEVONthink’s internal AI to retrieve data for Claude. It’s AI all the way down!
Could you please explain that? I might have missed something. But as I’m also working with (or rather experimenting with) Hermes Agent (and AgentZero), I’ve already been wondering how I might manage to integrate it with DEVONthink.
It’s magic: just a moment after I wrote that question, I received an update notification for Devonthink:“DEVONthink 4.3 Herschel introduces our MCP server:”
We should share our experiences of this new technology in a dedicated section of this forum.
my first question.
I don’t understand the difficulty.
Devonthink gave me the option to connect Hermes Agent, so I didn’t really have to do anything. It also had options for Codex, and Claude Code.
After that, just ask your agents to do what you want.
Or, once you have the MCP installed, just ask Hermes to set it up for you.
I don’t really go into any config settings anymore. Hermes does all the grunt work.
That wasn’t the case for me, as I use a Hemres Agent server on the LAN rather than a local installation on my computer.
Anyway, I’m sure I’ll find a solution.
I run it straight on my mac.. I keep backups separately, so I’m not worried about nefarious activity.
I think the biggest difference has been creating profiles.
Running Hermes gets expensive, so I need to disable skills, and keep only those that fit the scope of an agent. The agents are all ‘specialized’ so I have a coding, notes, and ‘general’ agent atm.
They each have their own ‘profiles’ which keep them in their lane. If you’re paranoid, you can even make it so that each only has access to one tool, for example emacs. I use the notes one with access only to emacs. One advantage of that is context window. It’s able to view a 500k token file because it doesn’t load the whole file, it only accesses a section of the file. I tell it to go to line 200 context 10 and it gets 10 lines above and below, and it’s able to edit sections instead of loading the whole file.
I’ve still got a lot to learn. It’s a whole new world for an old white man ![]()
In my experience, the Hermes agent is more cost-effective in a production environment than AgentZero, as Hermes uses better caching.