If you add a system property “Source”, and put the link there, and then do “Reload from source”, it already does something of that kind.
hey, so with WWDC2025 - can the local iOS ai be used with Anytype?
Any way to just talk to the selfhosted nodes directly via the api so no need to desktop clients?
I have automations etc but dont really use desktop so would love to just be able to run scheduled automations on just server side and let them sync to my phone plz
This is my only deal breaker for me to still use obsidian with git
just need to be able to remotely change my own data
@xrehpicx yes, we have plans for headless anytype, and already a prototype internally. also see my answer here: API and Integrations · Issue #19 · anyproto/roadmap · GitHub
Hey, I’m glad about the release of the API — it’s exciting news. I’d like to know if the Anytype team has plans for ‘more comprehensive plugins’? Something similar to Obsidian’s plugin system. What I mean is the ability to customize blocks (both logic and rendering), customize page types (currently there’s note, page — I’d love to expand these options through plugins), and customize app interfaces (like adding buttons), etc.
Following.
I NEED reoccurring tasks and then I am moving everything to Anytype and not looking back!
Headless Anytype - fine, that’s certainly a great, ambitious goal (and AI is sure to come around the corner too). But - isn’t Anytype still a “note taking app” - in beta?
It hurts me to read about goals that are so far away from the basics of working with a note taking app.
I remember my experience with Nimbus Note (now “FuseBase”), who one day decided they had to fly to the moon and turn a useful and well-functioning note taking app into a business collaboration software that claims to can do everything - and in the process destroyed their primary note taking app capabilities permanently. While of course being still far behind Notion, Coda and even ClickUp in terms of possible use in companies.
Are you guys now really starting to go after enterprise customers too, while Anytype definitely has its hands full being a solid note taking app for individuals?
We already have chats, but currently not even the web publishing functions are fully developed, not to mention the mobile apps for Android and iOS. I can only speak for the Android-App - there is really a lot to do. And not to mention the many requested non-collaborative functions that other comparable apps have long since implemented and which have often still not been realized despite a high number of votes. And not to mention the many existing and newly reported bugs on a daily basis.
And on top of that, we also have to deal with statements of co-founder @anton thinking out loud in a thread about ditching the whole thing altogether and giving it to the open source community and making a new app, only to then qualify and basically deny it because he’s seen the uncertainty it’s caused amoung users (I can provide the source link, but I think it is well known here).
So I think there are really more important topics at Anytype at the moment than “headless”.
I’m having problem testing the MCP server. I use Raycast chat as shown in the picture, just wanting to search for some of my objects. Am I doing something wrong here? @jannis
@dzlg hmm, you might wanna use a non-thinking model to better debug the tools it tries to call & which step breaks. that error isn’t anytype specific actually, rather looks a malformed json issue..
if the problem persists, you can cmd+k > copy log and share it privately, then i could take a look
@anton Can you confirm that AI assistants will not be integrated natively into Anytype? This is a real concern that needs clarification.
@BioLinua @VisualNotes we haven’t yet discussed a ‘store for plugins’ in detail within the team. there is still more work to be done on the api level, which i will address, to ensure the api is rock solid. that comes first; people can already build integrations to external tools, and after gathering more feedback, we can advance further.
customizing the app interface or running in-app plugins requires more than just changing logic in anytype-heart; it also needs adjustments to the desktop client. i don’t have any estimates on that at the moment
it’s actually a very useful thing, and it’s just around the corner. no worries—the whole team isn’t focused solely on that.
we’re running a business here, serving people and organizations willing to pay. unfortunately, only a small number are ready to pay for private note-taking alone. from day one, our ambition has always been to build decentralized communication software—and only now have our protocols matured enough to support that vision. we’ve never planned or aspired to simply build a private note-taking app.
i’m sorry if it wasn’t clear or if you missed my explanation. let me state it plainly: there are no plans to ditch anytype. instead, we’re considering ways to make the experience clearer and more streamlined by rebuilding the app—not creating a separate one.
i might be mistaken, but the api is one of the most awaited features, and the headless middleware—which isn’t complicated to build—is a crucial part of it.
This is great! First real reason for me to get an AI MCP client up and running, which I’ve done with Claude. The integration was quite smooth and I’ve managed to populate a Space with some objects (ingredients) based on a webshop.
Issue I’ve found: I can create objects, types etc with Claude, but I cannot actually update the Markdown body of a page. In this case I wanted Claude to update the internal links. It knows how to do this, but actually posting the updated markdown does not lead to any change.
Is this a known issue?
As per Claude:
I can see the update was processed (the
last_modified_datechanged to2025-06-13T09:01:39Z), but the markdown still shows the old content. Let me check if the update actually took effect by retrieving the object again:
API-get-objectI can see that the update attempt was registered (the last modified date changed), but the markdown content is still showing the original template format. This suggests that the AnyType API might have specific limitations or requirements for updating body content that I’m not meeting.
@mvschie yes, updating body isn’t possible yet, see here: API, Developer Portal & MCP! - #36 by jannis
Hello, I’m coming from the email. This is so cool! Is that API complete enough so I could (theorically) write my own front-end?
You can write your own front-end without API =)
I don’t know how it would be possible but that’s quite amazing. You guys are the best.
Nice! there are so many things i would love to connect to anytype but i am not a coder or know anything about coding.
Hi guys! Thanks for this valuable update. Want to explore API in the future. But right now I’m trying to figure out how to work with MCP locally. Since I’m on Win11 I’m testing clients from this list GitHub - punkpeye/awesome-mcp-clients: A collection of MCP clients., cursor.ai and VSCode extensions. Unfortunately I didn’t find a way to use it fully local: MCP client (app) + Ollama (model) + Anytype.
I found that it may work in cursor ai app with cloud based model, i.e. it can find and rename object “a” in a space “b”. But cursor can’t support Ollama models natively, so I have to use third-party software in order to make a bridge between cursor and ollama and I’m not sure that it may be fully local.
I’ve tried most of the free MCP clients and even though they successfully managed to work with ollama models and MCP client see the Anytype MCP server (i.e. toggle is on or marked with a green colour) it can see 29 API-… commands
But when I use ollama qwen3:14b model which can work with external tools, and ask it the same prompt as for the cloud based model, it doesn’t answer that it found object “a” in the space “b”, but ask something abstract instead or write some code for that object searching. I think that 14b model is smart enough to work with MCP tool, so I think there is something wrong, but I can’t figure out what I’m doing wrong here. If I ask the model, what tool it can use, in prints the API-… commands list.
I remember that behaviour in GitHub - nbonamy/witsy: Witsy: desktop AI assistant / universal MCP client and GitHub - nanbingxyz/5ire: 5ire is a cross-platform desktop AI assistant, MCP client. It compatible with major service providers, supports local knowledge base and tools via model context protocol servers .
does this mean one could add a way to add handwritten canvas feature into the Anytype its self?