Both give Claude context. They live in different layers.
Anthropic Projects let you create a workspace in Claude, upload files, and add project-specific instructions. The combination of Projects and Files is a clean way to scope a chunk of work and keep Claude grounded inside it.
Hjarni is one layer below. A persistent knowledge base your AI reads through MCP. Same notes work in Claude.ai, Claude Code, ChatGPT, Cursor, Codex, and any other MCP client. Anthropic's Projects scope context to Claude; Hjarni scopes context to you.
Project-scoped knowledge versus reusable memory
Anthropic Projects + Files is great when the work has clean edges and a deadline. You upload the relevant files, set project instructions, and Claude works inside that envelope until the project ends.
Useful knowledge usually leaks across project boundaries. Customer research informs multiple launches. Architecture decisions show up in many code reviews. Style guides apply everywhere. Re-uploading the same files into every new project is the part that gets repetitive.
Hjarni's bet is that knowledge should outlive any one project. Notes stay in the knowledge base. You spin up new conversations whenever you like; the context is already there.
Anthropic Projects scope context to one workspace inside Claude. Hjarni scopes context to you across every assistant.
A concrete workflow difference
You're preparing a product launch. With Anthropic Projects + Files, you create a Launch project, upload the brief, design notes, and research docs, and chat with Claude inside that workspace. Anything beyond that project is out of scope.
With Hjarni, the launch notes live alongside the rest of your product knowledge: prior launches, retros, customer interviews, decisions. You ask Claude through MCP to draft the launch plan using all of it, then write the plan back to the same folder. Tomorrow you switch to Cursor for the engineering work and the same notes are still there.
Read-only uploads versus memory that updates itself
Files inside an Anthropic Project are context Claude can read, not a surface it can write to. The assistant answers from what you uploaded, but it cannot edit those files. When a fact changes, the loop is manual: regenerate the text, delete the stale file, and re-upload the new version. The knowledge also stays inside Claude, so nothing you put there is reusable from another assistant.
Hjarni keeps the same knowledge reusable across every assistant, and through the MCP server an assistant can create or update a note in a single step, so the memory updates itself without a download-and-re-upload cycle. The case is not that Anthropic Files cannot hold knowledge. It is that write-back is a capability Hjarni exposes and Files does not, and Hjarni's edits land as visible, named, revertable notes you can read and roll back, not a silent overwrite.
When Anthropic Projects + Files is the better fit
If Claude is your only assistant, work breaks cleanly into discrete projects, and uploading files into a focused workspace matches how you think, Anthropic's stack is a strong default. Staying inside one ecosystem keeps the setup short.
When Hjarni starts to make more sense
The case for Hjarni gets stronger as soon as you use more than one assistant, or your knowledge needs to outlive any one project. Teams who share context across people and tools. Builders who use Claude and Cursor and Codex in the same week. Anyone who's tired of re-uploading the same set of files into new projects.
It also gets stronger when you want folder-level AI instructions (a writing folder behaves differently from a runbook folder) rather than one project-level instruction blob per workspace.
Pricing and portability
Anthropic Projects and Files sit on top of paid Claude plans. Hjarni's free tier includes the MCP server, and Pro starts at 9 EUR per month. Hjarni notes export as a Markdown ZIP at any time, with no lock-in. Files inside Anthropic Projects live in that project and are retrieved per-file from the Anthropic surface.
Often used together
Many teams use both. Anthropic Projects for in-flight launches, Hjarni for long-term knowledge that every assistant should read. The MCP server makes that combination clean: Claude in a Project can still call Hjarni for the notes the project does not contain.