---
title: "01 Personal Access Tokens"
canonical_url: https://mdflow.cz/share/c/h0L7MvX5Cy5sST3qPnApqHOmrQ36JTVy169qJMtDMya9Kc30zm04cg9OuPSSPwvj/bdefa552-e9b6-4382-a479-91615474cf69
md_url: https://mdflow.cz/share/c/h0L7MvX5Cy5sST3qPnApqHOmrQ36JTVy169qJMtDMya9Kc30zm04cg9OuPSSPwvj/bdefa552-e9b6-4382-a479-91615474cf69.md
visibility: public
---

# Personal Access Tokens

A Personal Access Token authenticates scripts, tools and AI agents against your MDflow account.

**Creating and using tokens requires Pro.**

## Creating one

1. Go to **[mdflow.cz/settings](https://mdflow.cz/settings)**.
2. Create a **named** token — name it after what will use it (`Web Clipper — laptop`, `n8n production`, `research script`).
3. **Copy it immediately.** A token is shown **once**, at creation, and never again.

Tokens look like `mdf_…`.

If you lose a token, you cannot retrieve it. Revoke it and create another.

## What a token can do

This is the section to read properly before you paste a token anywhere.

> A Personal Access Token grants **read and write access to the token owner's entire account**: reading, creating, updating, moving and deleting documents and folders, and managing sharing.

Concretely, it is **not** scoped by:

- **Workspace** — a token sees *every* workspace, including private ones. Separating content into workspaces does not sandbox a token.
- **Folder.**
- **Permission level** — there is no read-only token.
- **Tool** — a token is not restricted to the app you made it for.

There is one grant, and it is total.

## What a token cannot do

Two meaningful limits:

- **It cannot read version history.** History is private to the owner and is not exposed through the API or MCP server.
- **It cannot read encrypted documents.** It fetches ciphertext. Without the password — which never reaches the server — there is nothing to read.

If there is content you want to keep away from your own agents, encryption is the mechanism. Workspaces are not.

## Managing tokens

For each token you can review:

- its **name**,
- its **creation date**,
- its **last-used date**,
- its **current status**.

The **last-used date** is the useful one. A token you do not recognize, with recent activity, is worth revoking immediately.

## Revoking

Revoke a token and it stops working **immediately and permanently**.

**Revoked token records are retained** for account visibility and audit history — you keep the record of what existed and when it was used, rather than the evidence quietly disappearing.

## How tokens are stored

MDflow stores **token hashes, never the plaintext token**. Nobody at MDflow can read your token back — which is also why it can only be shown once.

## Practical advice

**One token per tool.** It costs nothing and it means you can revoke the clipper's access without breaking your n8n workflows, and vice versa. It also makes the last-used dates meaningful.

**Revoke on any doubt.** Old laptop, uninstalled extension, a token pasted into something you now regret. Revocation is free, immediate, and non-destructive to your data.

## What uses tokens

| Tool | Needs a token? |
| --- | --- |
| **HTTP API** | Yes |
| **Remote MCP server** | Yes — or OAuth, with no token to copy |
| **Local MCP server** | Yes |
| **Web Clipper** | Yes |
| **VS Code / Cursor extension** | Yes — or OAuth |
| **n8n node** | Yes |
| **iOS app** | **No** — it signs in normally and is not Pro-gated |

Where OAuth is available (Claude.ai, ChatGPT, and the editor extension), prefer it. There is no token to copy, store, or leak. See [04 Connecting Claude, ChatGPT and Other Clients](/doc/72aa97f1-fef9-4465-82fc-0a21ee327252).