Your AI assistant already knows a great deal about your current problem and nothing at all about the eight months of notes where you solved a version of it before.
The Notez MCP server closes that gap. Model Context Protocol is an open standard for connecting assistants to external tools and data, and Notez ships an MCP server that exposes your workspace over it.
Setup
1. Generate a key. In Notez, go to Settings → Integrations, find Notez MCP server, and click Generate MCP key. Copy the endpoint and the bearer token — the token is shown once.
2. Add it to your client. The Integrations page gives you a ready-to-paste command for Claude Code with your endpoint and token already filled in, plus JSON config for Claude Code, Cursor, and Windsurf with the correct file location for each. Claude (web, desktop and mobile), ChatGPT and VS Code don't use a key: they connect by signing in, from the Connect an AI app card on the same page.
Copy from that page rather than typing it by hand — the authorization header has to be exactly right or every call fails.
3. Restart your client and ask it something about your notes.
The tools
A key gives an assistant 31 tools: 13 that read and 18 that write. The table lists the ones it reaches for first, and MCP server setup lists them all.
Tool | What it does |
|---|---|
| Which workspace this key is scoped to |
| Search and list, with filters and pagination |
| One note, summary or full |
| Direct children of a note |
| The breadcrumb chain to the root |
| Every tag with note counts |
| Semantic search by meaning, up to 5 results |
| A kanban board with its cards grouped by column |
| Create a note, Markdown supported |
| Change part of a note and leave the rest untouched |
| Move cards between the columns of a board |
| Move a note to the Bin, or bring it back |
list_notes supports substring search, tag filters, board columns, custom property values, created and updated dates, note type, favorites-only, top-level-only, trash, ordering, and pagination — so an assistant can navigate a large workspace without pulling everything into context.
Two limits worth knowing up front
It can't delete structure. A key reads and writes: an assistant creates, edits, renames, tags and moves notes, deletes them to the Bin and restores them. No tool deletes a tag, a property, an option or a board column, so removing those stays with you. The in-app AI chat runs the same tools, apart from the two for file uploads.
Keys are workspace-bound. A key generated in your Work workspace grants access to Work only. To expose a second workspace, switch to it and generate a separate key. Rotate any key from the same screen.
What this is actually good for
Ending a session with a written record. The most valuable moment in debugging is right after it works, and it is the moment you are least inclined to write anything down. "Create a note titled Postgres pool exhaustion with what we just figured out, tagged infra." Done, before you close the terminal.
Recovering prior context. "Search my notes for anything about rate limiting before we design this." Your assistant reads what past-you concluded instead of re-deriving it, possibly differently.
Meeting prep. "Find all my notes tagged acme from the last two months and summarize the open threads."
Working across the boundary. Notes created over MCP carry an MCP badge, so you can always see what your assistant wrote versus what you wrote.
On trust
An MCP key grants read and write access to a whole workspace. Treat it like any other credential: it is shown once, it lives in your client's config file, and it should be rotated if it leaks. Rotation is one click on the Integrations page and immediately invalidates the old key.
The workspace binding is the real safety property here. If you keep personal and work notes in separate workspaces — which is the recommended setup — then a key in a work project genuinely cannot reach anything personal. That is enforced server-side on every tool call, not by prompt instructions.
Continue reading
How to Give Claude and Cursor a Persistent Memory With MCP — one notes workspace as the memory every assistant reads and writes, and what belongs in it.
Your notes in Claude and ChatGPT, without an API key — adding Notez to Claude, ChatGPT or VS Code with one URL and a sign-in, and taking access back.
Claude Code can now put images in your notes: file uploads over MCP — how an assistant with a shell puts a screenshot into a note without sending it through the chat.