I am thinking of buying a pro license with Claude. Can I use this license to integrate Claude with Devonthink or do I need to pay for a separate API key?
This sounds more like a Claude question, but Claude Pro does not come with API access—you need to pay for it separately.
It’s not a stupid question at all.
Based on how their pricing is structured:
- If you want to use Claude for chatting, AI-driven smart actions, etc., you’ll need an API key.
- If you want to use DEVONthink’s MCP server, e.g., in Claude Desktop/Cowork/Code, you’ll need a Pro or Max license.
„Oops! That page doesn’t exist or is private“ Is it the right link?
I initially used Claude via a token-based API interface directly in DEVONthink. That connection is still active, but I hardly use it anymore, since — thanks to the MCP server in DT — I now use the Claude Pro version as chat, but above all with Cowork. Works perfectly. The advantage is that it runs in the background in parallel, and it’s much better for my old eyes, since it’s in a large window. On top of that, with the flat-rate costs, it’s noticeably cheaper for me than with the API. I can only recommend it.
Translate with Claude Sonnet 5
Thank you all for your answers. It is very helpful. I have a pro license of Devonthink 4. Do you need to install the desktop version of Claude in order to use it with the MCP server? What kind of access to Claude need? I am a bit reluctant to give Claude access to my personal computer.
The desktop app is indeed required. Access to DEVONthink’s MCP server is sufficient for this usage scenario, the Claude.app doesn’t need additional permissions. In DEVONthink’s it’s also possible to exclude databases, groups or documents (see Info > Generic inspector and File > Database Properties) from chat & MCP.
In addition to the settings within DEVONthink itself for restricting access to databases, groups, and individual files via the MCP server, you can also configure access and permissions within Claude Cowork or Claude Code, so that every access to the data in DT by Claude must be requested and approved. This is very secure, but correspondingly time-consuming as well.
Translate mit Claude
Or just use the Claude Code CLI wired up to your own local hosted model and DEVONthink. I haven’t connected DT to it yet, but that’s otherwise how I dodge the subscription fee and avoid giving big tech morons every byte of my data.