Claude Meets DEVONthink: An Assistant for My Research Library

[Bmosbacker]

Disclosure: After working on this today for a chapter in my book, I decided to share with you my experience. After doing this work with Claude, I had Claude provide the summary that appears below.

I have written here before about my writing and research workflow, which centers on DEVONthink as the archive for years of accumulated research in theology, leadership, philosophy, science, technology, education, and more. I recently connected Claude to DEVONthink, and the result has changed how I work with my own material. I want to describe how it works and why it matters, not how to set it up.

How It Works

The connection gives Claude direct access to my DEVONthink databases. From an ordinary Claude conversation, I can ask it to search my records, read a document, pull together my highlights and annotations, find documents related to one I am studying, or summarize a folder of research. Claude sends those requests to DEVONthink on my Mac, receives the results, and works with them in the conversation.

Two points deserve emphasis. First, my documents remain where they have always been. Nothing is uploaded wholesale, nothing is moved, and my database structure is untouched. Claude retrieves only what a given request requires. Second, I do not construct search syntax or export files for analysis. I ask in plain English, and the assistant does the searching and the assembling.

The Benefits

The principal benefit is that the answers come from my own library rather than from the open internet. When I ask for a summary of what I have gathered on a subject, the answer draws on sources I selected, read, and annotated over many years. I know where every claim came from because I put it there.

Second, the tedium of manual transfer is gone. The old pattern of locating a document, copying passages, and pasting them into a chat has ended. The assistant goes to the library itself.

Third, it recovers what I have forgotten. A database built over many decades contains far more than memory can hold. Asking Claude to draw connections across hundreds of records routinely turns up material I had not thought about in years, which is precisely what a research archive is for.

[FrMichaelFanous]

I am very interested to hear as you progress through this on a regular basis. This will probably bring my DEVONthink usage back to life instead of being a ‘dump-bucket’ of things that I want. The last 3-4 months I was just dumping things into the Global Inbox to ‘come back’ and organize.

[KVZ]

The DEVONthink / MCP / Claude setup is a significant advance for research and writing. Of course, it depends on our own ability to curate useful databases. I appreciate DEVONtechnologies’ attention to privacy – optional exclusion of our data from chat and the MCP with settings at the database, group, document, or data element level. I believe DEVONthink’s approach is far more appealing and rational than letting AI models loose on the entire contents of a personal computer, phone, cloud storage, etc.

Katie

[Yaakov]

DEVONthink is often dismissed as little more than a document warehouse, but I’m curious whether others are pushing it further than that. Does anyone here use it as their primary environment for note-taking, outlining, journaling, and PKM — an all-in-one thinking and writing tool, not just a place to store files?

I’d love to hear how you’re actually working in it day to day. Thanks in advance.

[kayle]

I, too, will be following this thread. I am a new user of DevonThink - trying to find my way around it. I am mostly using it for research into 2 documentary films I am working on. Being able to put hundreds of Documents in and have Claude search only them for connections sounds ideal.

It has a pretty steep learning curve for me, so still looking for tips and best practices and I navigate putting info in.

[KVZ]

@Yaakov

DEVONthink is often dismissed as little more than a document warehouse, but I’m curious whether others are pushing it further than that. Does anyone here use it as their primary environment for note-taking, outlining, journaling, and PKM — an all-in-one thinking and writing tool, not just a place to store files?

There’s a vast amount of territory between “document warehouse” and “all-in-one thinking and writing tool”, and you’ll find users all across the landscape of that territory. I’ve used DEVONthink since close to its original release, and have never used it as an all-in-one tool. I’m very suspicious of that use case since I prefer to use the best tool for a task and not all-in-one tools.

My case: I have 8 databases continually open all day every day in DEVONthink. They range from strictly-personal to work-related. My work these days is research, so I have four research databases for four different disciplines. I have a database for home and family matters, and two for working on smaller projects. Other than household record keeping (its own database) I tend to be project oriented. I keep research artifacts, artifacts I create, and analytical / notes.

I never create documents other than small scraps in DEVONthink. I store creations there but use the best tool for that particular document. (Which exact softwares I use is irrelevant to this thread). I suspect what I’m doing is probably similar to what @Bmosbacker is doing. The databases are the source material for what I write or the notes that inform my writing.

The best place to discuss use cases is the DEVONthink forum, which goes back years and is broad and deep.

Katie

[Bmosbacker]

@KVZ

I suspect what I’m doing is probably similar to what @Bmosbacker is doing. The databases are the source material for what I write or the notes that inform my writing.

Yes, that is precisely how I use DT. Although I keep all research material and my Kindle highlights in DT, I keep all my personal reference material (career and household stuff) in Apple Notes. Once every six months or so, I import my Apple Notes into an Apple Notes archive database in DT.

[Bmosbacker]

@Yaakov

Does anyone here use it as their primary environment for note-taking

I have tried. In my “ideal” workflow, I would use DT for all my notes, outlines, research, and more. Unfortunately, DT’s mobile app, though much improved recently, is still a poor cousin to the Mac version. I take a lot of notes on the iPad, making feature parity important to my workflow.

[krocnyc]

@Yaakov

Does anyone here use it as their primary environment for note-taking, outlining, journaling, and PKM

I tried, but I never found DEVONthink to be a congenial space for writing, reading, or working with documents for any extended period of time.

I also don’t import my documents into DEVONthink: I store them in Finder folders and index them in DEVONthink. That way, I can point DEVONthink, Obsidian, Typora, and whatever other app might be appropriate at the same collection of folders. If I need to link to a document from, say, NotePlan, I can create the link in DEVONthink even though the file has been indexed there rather than imported. (A note about indexing: you do have to be diligent about refreshing the index in DEVONthink.)

[MacDependent]

I’ve been wanting to love DT and use it more in my workflow but it never really neatly fit for me. I still use it occasionally to search for specific files so I think this new integration with Claude will be super helpful to me. My DT database is indexed. Does anyone know if DT will still work with Claude if database is indexed?

[Joost]

I have DT databases with patents, datasheets, research papers, competitive marketing materials etc collected since i started using dt (around 2012 I believe). Many many docs! Since december last year i started working with them through (my own) MCP and my workflow is now drastically different. I mostly work in claude code, searching, summarizing, combining technologies/findings all go through CC. The mcp also can direct CC to external databases (webmd etc) if it finds no or few documents of the topic at hand in my db and then can download it.

DT is many tools wrapped up as one and thereby provides a cohesive interaction layer for us humans but i am beginning to see how dt is no longer necessary for some aspects/users.

One element i feel is a must if you work research through CC is a ticket system. A ticket specifies the “problem to research or work on” and CC maintains its findings in that same ticket. That then allows you to verify outcomes or track its reasoning.

Exciting times…

[Amontillado]

@Yaakov

DEVONthink is often dismissed as little more than a document warehouse, but I’m curious whether others are pushing it further than that.

The flexibility of alternate grouping by using tags, which are themselves really just groups, is nice.

I use replication when a document really should appear in more than one place. That’s not the same as symbolic linking. Each instance of a replicant is the original file, not a subordinate, and replicants don’t go away until every instance has been deleted.

Annotation files are really cool. That’s where you attach a second text (Markdown or RTF) file to one you don’t want to modify and make your notes in the annotation.

Leveraging grouping, tagging, linking, replication, and transclusion offer some nice ways to emerge order from a large number of files. The search and see-also functions really help, and the AI support for those into that kind of thing looks strong.

It’s a tool that does nice things for me when I’m working with facts and documents.

[krocnyc]

@MacDependent

Does anyone know if DT will still work with Claude if database is indexed?

Yes, it does! I find using the MCP fits in better with my workflow than using Claude IN DEVONthink or pointing Claude itself at the relevant Finder folders.

By way of example, here’s a quick query I just ran from Claude Cowork. Note that my prompt is pretty basic; I’m usually more explicit about what I want Claude to extract from my database, how I’d like it formatted, and what kind of artifact I’d like to to generate (e.g., a markdown file, a spreadsheet, etc.)

Me:
Please search my DEVONthink Reference database for all articles that discuss Sebastião Salgado. What is the main argument of each one?

Claude:
Many hits are passing mentions in books. The actual articles are the markdown clippings. Let me read the strongest candidates.

Let me verify whether the other notable hits substantively discuss Salgado or just mention him in passing.The two New Yorker clippings are the clear articles.

Let me check whether the Kember essay (an article in The Photography Reader) substantively discusses Salgado.

Two articles in your Reference database actually discuss Salgado. The other hits (the La Grange, Kember, Berger, and other books) only reference him in passing or in a footnote, not as a subject. Notably, the two articles are in direct dialogue with each other.

Chris Wiley, “Sebastião Salgado’s View of Humanity” (The New Yorker, 2025) — an obituary/appreciation. Its argument is a defense of Salgado against the long-standing charge that he aestheticizes suffering. Wiley contends that Salgado’s “operatic,” gorgeous style is not a betrayal of his subjects but a “potent enhancement of his act of bearing witness” — that viewers are sophisticated enough to separate content from form, and that his beauty asserts the dignity and humanity of the poor, workers, and refugees rather than reducing them to victims. He traces the arc from Workers and Exodus to Genesis and the Instituto Terra rewilding project, framing Salgado’s career as the end of an era of ambitious essayistic photojournalism.

Ingrid Sischy, “Good Intentions” (The New Yorker, 1991) — the original critique Wiley is answering. Its argument is the opposite: that Salgado’s formal beauty is a moral failure. By rendering famine, labor, and suffering as gorgeous, religiously-loaded, symbol-laden compositions, he “aestheticizes tragedy,” which “is the fastest way to anesthetize the feelings of those who are witnessing it. Beauty is a call to admiration, not to action.” She argues his work is heavy-handed, derivative, and self-aggrandizing, that the cult of appreciation around him ends up admiring the photographer rather than helping his subjects, and that his images reinforce viewer passivity rather than driving change — a gap between good intentions and results.

Read together, they form a point/counterpoint on the central question in documentary photography: whether making suffering beautiful honors its subjects or neutralizes the viewer’s response.

Sources:

• Chris Wiley — Sebastião Salgado’s View of Humanity
• Ingrid Sischy — Good Intentions

[cornchip]

Finding some value in mixing LLMs and DEVONthink as well. MCP servers or ad hoc AppleScript are definitely the right abstraction for LLMs. I try to use LLMs to do things that are good ideas, but tedious, and require a bit of decision-making, i.e. not a good candidate for a script. Finding items/groups that should be replicated, deleting attempts to start a collection that went nowhere, identifying islands of items that could break out into their own database are some examples.

I’ve done almost zero generation of text with it, either through an MCP or the built-in tools.

@Yaakov

Does anyone here use it as their primary environment for note-taking, outlining, journaling, and PKM — an all-in-one thinking and writing tool, not just a place to store files?

Yes; I don’t go whole hog on PKM, though. But, for example, at work I hit a keyboard shortcut to start yyyy-mm-dd dated notes. Usually I start in the folder for the project. I take a lot of these.

For more research-y projects, small/atomic notes work pretty well and if you add a note that mentions B/C/D in a note about A, it helps the classifier to make connections for other documents. I haven’t done it but this could be a good use of an LLM, to generate individual fact files that glue other documents together.

I use the markdown editor and pushed the look of it to be as similar to iA Writer as possible. Still could be better.

Small plain text or markdown notes, longer markdown writing, and wiki editing all work fairly well.

[DEVONtech_Jim]

@MacDependent

Does anyone know if DT will still work with Claude if database is indexed?

There is no specific issue with indexed documents and Claude with DEVONthink’s MCP server.

[DEVONtech_Jim]

@krocnyc

Me:
Please search my DEVONthink Reference database for all articles that discuss Sebastião Salgado. What is the main argument of each one?

• How large is this database?
• Did Claude start with an initial database search?

[krocnyc]

@DEVONtech_Jim

• How large is this database?
• Did Claude start with an initial database search?

1. It’s not my largest: 1,180 items; 6.6GB; 51.2 million words, 336 thousand unique words.
2. Do you mean a search inside the particular DB I pointed Claude at, or a search for DEVONthink databases? If the former, I assume so.

[DEVONtech_Jim]

Thanks and I was just curious if it’s something you had noticed

[krocnyc]

Some further detail: the Cowork progress indicators show tool calls like “get_databases,” “search_records,” and “get_record_text” with a “D” icon next to them. It’s my understanding that the “D” indicates that the source was DEVONthink. If I open the “Result” capsule under “search_records,” it shows the following:

Request

{
“query”: “Salgado”,
“database_uuid”: “BB9BF456-FE43-40BF-8193-C2E79F43AC39”
}

Response

{“count”:9,“results”:[{“databaseUUID”:“BB9BF456-FE43-40BF-8193-C2E79F43AC39”,“uuid”:“2A635D69-5AA8-415A-B6DD-9E172593988A”, … etc …

This looks like Claude is searching the DB to me, but I’m new at this …

In any event, unless you open up the progress indicators for the detail, whatever is going on is unobtrusive.

[allan.galbraith]

Dr. Mosbacher,

Thank you for sharing this. I clipped it right into my DevonThink database on Claude. I’ve been really excited to find use cases for all the material I have filed over the years into DevonThink and this was very helpful.