So this comparison has a column the usual lists skip: hosting location. And a second look that's usually missing: what the tools can do with documents, not just typed text.

The comparison

Sorted by situation

The team writes wiki pages and lives in Jira: Confluence — the standard, sluggish but connectable to everything. The team wants it lighter: Nuclino (German company, EU-hosted, pleasantly reduced) or Slite. Full control, own servers, IT on hand: BookStack — open source, proven, free; maintenance is the price. Individual with text knowledge: Obsidian or Capacities — the detailed comparison is here.

The professional lens: law firms, tax offices, libraries, archives

Here the usual recommendations flip, for three reasons:

1. The knowledge is documents. A wiki doesn't help when the answer sits in a scanned notice from 2019. What's needed is text recognition and search inside the document — the wiki camp (Confluence, Notion, Nuclino, Slite, BookStack) indexes attachments superficially at best.
2. Data location is a duty, not a preference. Client and personal data on US servers is simply not an option for many firms and public bodies. That filters out Notion and makes the EU column above the first sorting question.
3. Answers must be verifiable. "It's somewhere in the file" doesn't cut it — you need the source: which document, which page.

That profile is exactly what Memexa is built for: documents and scans run through text recognition, get linked in a knowledge graph, and every answer names document and page. On top comes the part classic knowledge management software doesn't have yet: the knowledge is usable by Claude or ChatGPT via MCP — "what did we decide about client X's depreciation in 2023?" becomes a question to the AI, with a cited source. EU-hosted, GDPR details here.

To be fair: Memexa is not a team wiki with page permissions and editing workflows. If you need both, combine them — wiki for the processes, document memory for the files.

Frequently asked questions

What does it cost to start?

BookStack is free (plus a server), Nuclino and Slite start with free tiers, Confluence and Notion likewise. Memexa has a permanently free plan; paid plans start at €12/month. For a tax office the more relevant math is usually search time: one unfound source costs more than any monthly license.

Can you do knowledge management without the cloud?

Yes — BookStack (your server) and Obsidian (your device) answer the question fully locally. The price is maintenance, or doing without team access and document text recognition.

How do you start without overreaching?

With the existing stock, not with structure: make the documents you already have searchable first, then document processes. A searchable archive gets used from day one — an empty wiki doesn't.

The route below gets everything out, keeps the text notes in open Markdown, and keeps the scans searchable — with page-level sources on top.

Step 1: Export everything from Evernote

In the Evernote desktop app, select a notebook and export it as ENEX (.enex) — repeat per notebook to keep the structure. ENEX is Evernote's XML format and contains your notes and their attachments. Do this on desktop; the web app doesn't offer full export.

Two practical notes: export notebooks separately (a single giant file makes the next step slower and errors harder to find), and check the file sizes — a notebook that's suspiciously small usually means offline notes weren't synced.

Step 2: Convert with the Obsidian Importer

The free, open-source Obsidian Importer plugin reads ENEX and writes clean Markdown — one file per note, attachments in a folder, links preserved. Install Obsidian (free), add the Importer plugin, point it at your .enex files.

This step is worth doing even if you don't plan to keep Obsidian, because it converts your archive from a proprietary XML into the one format every tool on earth can read. The same importer also handles Notion, OneNote, Apple Notes, Bear, Roam and Google Keep — this guide's route works for all of them.

If Markdown files on your machine are all you wanted, you're done here, and honestly: for pure text collections that's a fine place to stop.

Step 3: The scans — where most routes end

Now the honest part. Your notes are Markdown; your scans are image and PDF attachments sitting in a folder. Obsidian will show them, but it can't read them — file-based tools index text, not pixels. In the Mac Power Users forum, an Evernote emigrant put it plainly: Obsidian is "good at text only/minimal images", and getting receipts and scans in is "clunky". That's not a knock on Obsidian; it's just not what it's for.

So the receipts and contracts that made Evernote your external memory become file names. Searchable by title, mute inside.

Step 4: Make the scans searchable again

This is the step Memexa exists for. Zip the vault folder from step 2 and upload it — notes become linked entries in a knowledge graph (wikilinks preserved), and every attachment runs through OCR. The scanned contract is readable again, and better than before: search hits cite the source down to the page, and any MCP client — Claude, ChatGPT, Cursor — can query it. The setup guide is here.

The chain in one line: Evernote → ENEX → Obsidian Importer → Markdown → Memexa. No step loses data, every step uses an open format, and you can stop at any rung and still be better off than inside Evernote.

And if you just want a new note app?

Fair question, short answers: Joplin is the closest free, open-source Evernote replacement (clipper included). Obsidian if you want linked Markdown notes and are happy text-first. Capacities if you want structure without building it, hosted in the EU. Our full comparison of nine tools covers the field — including where each one hosts your data.

Frequently asked questions

Does the OCR handle German documents? Handwriting?

Printed text in the usual European languages works well, including German umlauts and mixed-language documents. Handwriting recognition is honest-to-goodness harder; clear block writing often works, cursive is a lottery.

What about Evernote's tags and notebooks?

The Importer turns notebooks into folders and tags into Obsidian-style tags; Memexa's import maps both into the graph. Nothing is silently dropped.

How long does the whole migration take?

The Evernote export is the slow part (minutes to hours, depending on archive size). Importer and Memexa upload are minutes each. Plan an evening for a decade of notes.

Route 1: Share the vault as a folder

The most direct route, and it costs nothing: start Claude Code in the vault folder (or share the folder with Claude Desktop) and drop in a CLAUDE.md that explains the structure. Claude then reads and searches the files like any other project.

What it does well: For text notes on one machine this is hard to beat. No setup, no cost, and data only leaves your machine for the model request itself. You version changes with Git.

Where it ends: The access exists only on that machine — on your phone and in claude.ai the folder doesn't exist. Claude searches with file tools (grep), not semantically: it finds "notice period" when the text says "notice period", but not when the note talks about "contract duration". And a scan in the vault stays a file name.

Route 2: A local MCP server

The community maintains several Obsidian MCP servers (usually started via npx, configured with your vault path). Claude gets real tools — read note, create note, search vault — instead of raw file access.

What it does well: A cleaner interface than route 1, sometimes with better search. Creating and editing notes is structured. Everything stays local.

Where it ends: It remains DIY: install Node, maintain a config file, track updates. The server runs while your machine runs. It answers neither the multi-device question nor the scan question.

Route 3: The hosted knowledge graph

Import the vault into Memexa as a ZIP (wikilinks survive), then connect Claude via MCP — the three-minute guide is here. From then on every client reads the same state: Claude Desktop, claude.ai, Claude Code, and ChatGPT or Cursor too.

What it does well: Hybrid search (semantic + full-text) instead of grep. Documents and scans go through text recognition and become citable — source down to the page. Works on every device without anything running on your machine.

Where it ends: It's a hosted service — your data lives with a provider (in our case: EU, GDPR, Markdown export any time), and beyond a certain usage it costs money. If you work strictly "nothing leaves my machine", take route 1 or 2.

The table

The honest reading: on "where the data lives" and on cost, DIY wins — and if all you need is your own text notes on one machine, route 1 finishes the job. The two rows where the hosted route stands alone — scans and sources — only matter if your knowledge includes documents. But then they decide everything.

Frequently asked questions

Do I have to give up Obsidian?

No. The usual setup is a division of labor: Obsidian stays the writing tool, the Memexa import runs periodically, and Claude works on the graph. Wikilinks survive the import as real edges.

What happens to my plugins and templates?

They stay in Obsidian and keep working — the import reads the Markdown files, not the Obsidian configuration. Dataview queries and similar plugin syntax arrive as text, not as executed queries.

Does this work with ChatGPT instead of Claude?

Yes — routes 2 and 3 speak MCP, which ChatGPT supports too. Route 1 is Claude-specific (folder sharing + CLAUDE.md).

One column you won't find in the other lists: where each tool hosts your data. In 2026 that question decides contracts, and most comparisons skip it entirely.

The comparison

Sorted by the problem you're solving

"Obsidian got fiddly — I want simple notes that sync" → Joplin

Open source, syncs through your own choice of backend (including your own server), decent clipper. It won't win design awards, and the linking is weaker than Obsidian's — but it removes the plugin maintenance that drives some people away.

"I think in outlines, not pages" → Logseq

Every line is a block, every block linkable, journal first. Open source and local like Obsidian; the switch is about thinking style, not features. If you write long-form documents, you'll fight it.

"I want local-first, but friendlier" → Anytype

Objects instead of files, encryption and sync built in and free. Younger project, smaller ecosystem, export less battle-tested than plain Markdown folders — but the most credible "Obsidian comfort upgrade" in the local-first camp.

"I want structure without building it myself" → Capacities

Typed objects (meetings, people, books) give you the structure Obsidian users assemble from plugins — out of the box, hosted in the EU. It's a personal tool; teams and heavy automation are not the point.

"My team needs it too" → Notion

The pragmatic answer for shared wikis, databases and project boards. You trade local files and open formats for collaboration — and your workspace lives on US servers, which for some teams settles the question either way.

"My notes should feed my AI — including the documents" → Memexa

This is the category where we actually compete, so said plainly: Memexa is not an Obsidian replacement for writing linked text notes — Obsidian is better at that. Memexa solves the part the file-based camp is honest about struggling with: PDFs, receipts, scans and mail. Uploads run through OCR, land in a knowledge graph, and Claude, ChatGPT or any MCP client can search them and cite the page. Obsidian vaults import as ZIP with wikilinks preserved, so the two work in sequence, not in competition. How it works.

When you should just keep Obsidian

• Your knowledge is text you write yourself, and the vault structure works for you.
• You value local files and an open format above sync convenience — that combination is still Obsidian's home turf.
• The plugin ecosystem solves problems you actually have (not ones you might have someday).

Switching note tools costs weeks of muscle memory. Do it for a problem you have today — for text-only knowledge, Obsidian plus discipline beats most alternatives on this list.

Frequently asked questions

Can I use Obsidian and an AI memory together?

Yes — that's the most common setup among our users: Obsidian for writing, Memexa as the layer AI clients read via MCP. Vault import preserves your wikilinks.

Which alternatives are open source?

Logseq, Joplin and Anytype. Obsidian's app is closed, but your data stays in open Markdown files — in practice that's the lock-in question that matters.

Which host in the EU?

Capacities and Memexa host in the EU; the local-first tools (Obsidian, Logseq, Joplin, Anytype) keep data on your device, which sidesteps the question entirely.

What MCP is

MCP (Model Context Protocol) is the open standard Claude uses to talk to external services. An MCP server provides Claude with tools such as "search my notes" or "read this document", and Claude calls them on its own when a question fits. More background is in the MCP fundamentals article; this paragraph is enough for setup.

Prerequisites

• A Memexa account (create one free, the Free plan is enough)
• Claude Desktop or Claude Code

You don't need API keys or config files. Sign-in happens through OAuth in the browser, and the address is the same for every client:

https://brain.memexa.ai/mcp

Connect Claude Desktop

1. Open Settings in Claude Desktop (gear icon, bottom left).
2. Choose the Connectors tab.
3. Click Add custom connector.
4. Paste https://brain.memexa.ai/mcp as the URL and confirm.
5. Sign in with your Memexa account in the browser window that opens. The sign-in is permanent.

Afterwards, "memexa" appears in your connector list.

Connect Claude Code

One command in the terminal:

claude mcp add --transport http --scope user memexa https://brain.memexa.ai/mcp

--scope user makes the connection available in all projects, not just the current folder. On first access the browser opens for sign-in. After that the connection runs in the background.

Check that it works

Ask Claude a question only your memory can answer:

"What do you know about me?"

If Claude calls tools like search_brain or list_recent and answers with content from your Memexa, the connection is up. In Claude Code, /mcp shows the status of every connected server.

What Claude can do afterwards

Claude now remembers things across sessions: say "Remember: release moved to the 14th" and the note is still there next week, in ChatGPT or Cursor too if you connect those as well. Uploaded PDFs and scans go through text recognition automatically; ask Claude about one and the answer names the source, such as "contract, page 4". New notes also link themselves to existing ones, so Claude can navigate the graph from one piece of information to the next.

Common problems

The connector shows up, but Claude doesn't use it

Claude only calls tools when the question sounds like it needs them. Be explicit at first: "Check my Memexa for …". After the first few hits, Claude starts using the memory unprompted.

Sign-in fails or hangs

Usually a browser issue: disable the pop-up blocker, then remove and re-add the connector. There is nothing to enter in Claude itself; sign-in happens entirely in the browser.

"Server disconnected" in Claude Code

Run claude mcp list. If memexa shows "failed", claude mcp remove memexa followed by the claude mcp add from above almost always fixes it. If the error persists, check whether a corporate firewall filters outgoing HTTPS connections.

Claude can't find a document

Freshly uploaded documents need a moment for text recognition and linking. You can see the status in the dashboard. Once the document is marked as processed, Claude finds it.

The protocol behind them — MCP, the open standard AI clients use to talk to external services — has its own article. This one stays on the practical side: the server.

What an MCP server does

An MCP server is a service that offers tools to an AI. Each tool is a named capability with a clear job: search my notes, read this file, create a GitHub issue. The client — Claude, ChatGPT, Cursor — receives the list of tools when it connects, and decides on its own when to call which one.

That's the whole trick. The server doesn't need to understand AI; it just answers requests. The model doesn't need to understand the service; it only sees described tools. Which is why the same server works with every client that speaks the protocol.

Which servers exist

The directories are full, but almost everything falls into four groups:

• Service connectors. GitHub, Notion, Slack, Google Drive, databases. The server translates between the AI and the service's API. Useful when you already work there.
• System tools. File system, terminal, browser automation (Playwright, for instance). Mostly interesting for developers building AI agents or letting AI write code.
• Search and retrieval. Web search, documentation lookup, scraping. They give the model current information that wasn't in its training data.
• Memory servers. They store what you tell your AI and make it findable in later sessions. The most interesting group if repeating your context forever annoys you — and the one with the biggest quality differences.

For everyday use you don't need ten servers. Most people are fine with one or two: one for work (GitHub, Notion) and a memory.

How to connect a server

Setup follows the same pattern in every client — enter the address, sign in once:

• Claude Desktop and ChatGPT: Settings → "Connectors", paste the server's address. Sign-in opens in the browser.
• Claude Code: one terminal command, e.g. claude mcp add --transport http memexa https://brain.memexa.ai/mcp.
• Cursor: one line in the MCP config file.

After that, the server shows up in the client's tool list and the AI uses it on its own whenever a question calls for it. What this looks like for Memexa is on the setup page — with copy-paste configuration for every client.

Local or hosted?

Many servers in the directories are small programs you start yourself — via npx in a terminal, configured in a JSON file. That's fine for developers and the most common hurdle for everyone else: the server only runs while your machine runs, you handle updates yourself, and on your phone it doesn't exist.

Hosted servers invert this. They run as a service on the web, you connect with a URL and a browser sign-in (OAuth), and the same state is reachable from every device and every client. In exchange, you hand your data to someone else — which brings us to the real question.

What to look for when choosing

The protocol is open; the quality behind it is not regulated. Three questions separate the solid servers from the risky ones:

1. Where does the data live? With local servers, on your machine. With hosted ones: which country, under which law? A server that doesn't answer this has already answered it.
2. What is the server allowed to do? A tool that can delete files or send mail deserves more suspicion than one that only reads. Don't connect more permissions than the task needs.
3. Can you leave? If the service disappears or you want to switch: is there a complete export in an open format?

Memexa's answers: EU hosting, only you, one-click Markdown export.

Frequently asked questions

Do I need to know how to code?

For hosted servers with OAuth sign-in: no — entering an address and signing in is all. Local servers from the community directories usually assume a terminal doesn't scare you.

How many servers can I connect at once?

Technically many; practically, restraint pays off. Every server makes the tool list longer, and the model walks it on every request — too many similar tools make its choices worse, not better.

What separates memory servers from each other?

Most are scratchpads: the AI stores text snippets it deems worth keeping and searches them later. That already helps. A hosted memory like Memexa goes further — you also throw in PDFs, scans and mail, everything is read, linked, and returned with its source, down to the page number. The difference shows three months later, when you want to know where a fact came from.

That's the problem behind an unassuming acronym you now see everywhere: MCP.

The short answer

MCP stands for Model Context Protocol — an open standard that defines how AI applications talk to external data sources and tools. Anthropic published it in late 2024 and deliberately built it as an open protocol rather than an exclusive feature. By now virtually every relevant AI client supports it: Claude, ChatGPT, Cursor, and many more.

The common analogy holds up: MCP is the USB-C port for AI. Before USB-C, every device needed its own cable. Before MCP, every connection between an AI and a data source needed its own hand-built integration. With MCP there is one plug that fits everywhere: a service offers its capabilities once as an "MCP server" — and every AI client that speaks the protocol can use them.

How it works in practice

Behind the protocol sits a simple division of roles:

• The client is the AI application you talk to — Claude, ChatGPT, Cursor.
• The server is a service that can do or knows something — it offers the AI "tools": search my data, save this note, read this document.
• You connect the two once, usually with an address and a short browser sign-in.

From then on the important part happens automatically: when you ask your AI a question that needs external knowledge, it calls the right tool, gets the result back, and works it into its answer. All you notice is that the answer is suddenly correct.

What does this have to do with memory?

Out of the box, every AI session is an island. The built-in memory features of ChatGPT and Claude help to a point — but they belong to the respective tool: what ChatGPT has learned about you stays in ChatGPT. Switch models and the memory stays behind.

MCP inverts that relationship. If your memory is its own MCP server, it no longer belongs to any client — it belongs to you. Every AI you connect reads and writes the same body of knowledge:

You tell Claude on Tuesday: "Remember: release moved to the 14th." You ask ChatGPT on Friday: "When was the release again?" — and get the answer. With the source.

That's the difference between a memory feature and memory as its own layer: models come and go, the knowledge stays in one place — searchable, linked, and exportable.

Do I need technical skills for this?

Fewer than the word "protocol" suggests. For most clients, setup takes minutes and always follows the same pattern: enter the MCP server's address, sign in once in the browser, done. In Claude Desktop and ChatGPT the menu is called "Connectors", in Cursor it's one line in a config file, in Claude Code a single command.

For Memexa it looks like this — including copy-paste configuration for every client.

Frequently asked questions

Does MCP cost anything?

The protocol itself, nothing — it's an open standard. Costs can appear on two sides: some clients only allow custom connectors on paid plans (ChatGPT, for instance), and the MCP server you connect may be a paid service.

Is it secure?

MCP itself only defines the language between client and server — how safe your data is depends on the service behind it: where it hosts, how it secures access, what it does with your content. The questions to ask any MCP service: Where does the data live? Who can read it? Can I leave any time? (Our answers: EU, only you, yes — as Markdown.)

Which clients support MCP?

As of today: Claude (Desktop, Code, Web), ChatGPT, Cursor, and a fast-growing field of other tools. Since the standard is open, practically every new AI application ships with MCP support — which means an MCP-based memory also works with the clients that don't exist yet.

There are many MCP servers — what sets them apart?

Hundreds. Most memory servers are developer tools: locally installed, self-maintained, storing whatever the AI deems worth keeping. A hosted memory like Memexa sits one layer up: you also throw in documents, PDFs and emails, everything is read via text recognition, linked — and every answer cites its source, down to the page number.