Docs site for your n8n workflows, written by GPT

Docs site for your n8n workflows, written by GPT

An n8n workflow that serves a small documentation site about the workflows on your own n8n instance. A page lists every
workflow with its status, dates and node counts. Open one workflow's docs and the site writes a description of it with
GPT, draws a Mermaid chart of its nodes and saves the page as a Markdown file. You can edit a page in a live editor and
write it again.

What is inside

• workflow.json: the workflow as n8n exports it, sixty nodes. Twelve are sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The model was gpt-4-turbo, which OpenAI shut down on 23 October 2026. OpenAI Chat Model now asks for
  gpt-5.6-sol, the replacement OpenAI names. The temperature of 0.2 is gone: the model's page does not list a
  temperature setting, and a reasoning model may not accept one.
• CONFIG pointed project_path at a folder inside n8n's own .n8n directory, which n8n blocks the file nodes from
  by default. It now holds the placeholder <YOUR_DOCS_FOLDER>.
• The docs branch of md files accepted any file name that contained docs_. A name such as docs_x/../../y.md could
  then point outside the docs folder. It now takes only names that start with docs_ and hold no
  /, \ or ...
• Both webhooks were open to everyone. Each now has an IP allowlist holding the placeholder <YOUR_IP_ADDRESS>, so
  nobody can use the site until you put your own address there.
• The docsify webhook used a fixed id as its path, copied from one webhook in the author's n8n. The path is now
  docsify, and you set it to match your own instance, as the steps below say.
• The Mermaid library was loaded without a version, and Docsify as version 4. They now load Mermaid 12.2.0 and Docsify
  4.13.1.
• The prompt asked for a key called workflow_desription, while the parser expects workflow_description. It uses the
  second spelling now.
• The links to the OpenAI and n8n credentials and the ids of two webhooks are gone.

Before you start

• A self-hosted n8n that you can restart or reconfigure. The workflow writes Markdown files to a folder, runs one shell
  command and reads environment variables.
• n8n's API on, with an API key.
• An OpenAI API key.
• A folder that n8n may write to. n8n limits its file nodes to ~/.n8n-files unless N8N_RESTRICT_FILE_ACCESS_TO names
  other folders, and it blocks the .n8n folder itself.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. In CONFIG, replace <YOUR_DOCS_FOLDER> with the folder's full path. If N8N_PROTOCOL and N8N_HOST are not set
   for n8n, replace instance_url with your n8n's address.
3. From n8n 2.0, the Execute Command node is excluded by default. Either remove n8n-nodes-base.executeCommand from
   NODES_EXCLUDE, or disable the node mkdir and create the folder yourself. A disabled node passes its data on.
4. Add an n8n API credential, with your instance's address followed by /api/v1 and your key. Select it in
   Get All Workflows, Get Workflow tags and Fetch Single Workflow1.
5. Add an OpenAI credential and select it in OpenAI Chat Model.
6. Save. Open single workflow and read the id in its production address, the part between /webhook/ and /:file.
   Open docsify and replace its path docsify with that id.
7. In the IP Allowlist option of docsify and single workflow, replace <YOUR_IP_ADDRESS> with the address or
   addresses that may use the site.
8. Publish the workflow and open <YOUR_N8N_HOST>/webhook/ followed by that id.

How a request runs

1. docsify answers the first visit with Main Page, a Docsify app. The app then asks single workflow for the files
   it needs, with the webhook's path as its base.
2. README.md is the overview. Get All Workflows fetches the workflows, Sort-workflows puts the newest edit first
   and Fill Workflow Table writes one table row each, with links to view, edit or recreate its docs.
3. summary.md is the left pane. Get Workflow tags and Workflow Tags turn the instance's tags into a list, and a
   click on one opens the overview for that tag.
4. A file named docs_<workflow id>.md is one workflow's page. doc action reads the action in the address. For view
   and edit, mkdir makes the folder and Load Doc File looks for the saved page.
5. With no saved page, Fetch Single Workflow1 gets the workflow, Generate Mermaid Chart draws its nodes as a
   flowchart and Basic LLM Chain asks GPT for a description and the nodes' settings. Generated Doc joins them and
   Save New Doc File writes the file.
6. Edit opens Edit Page, a live Markdown editor, and Save posts the text back to single workflow. Recreate writes
   the page again.

Change the result

• Where pages are kept: project_path in CONFIG.
• How GPT writes them: the system message in Basic LLM Chain, and the model in OpenAI Chat Model.
• The overview's columns: Fill Workflow Table and Instance overview.
• The look of the site: the Docsify theme in HTML_headers in CONFIG.
• The chart's shapes: the code in Generate Mermaid Chart.

Good to know

• Every first view of a workflow without a saved page, and every Recreate, is one paid OpenAI call that sends the
  workflow's nodes and settings to OpenAI. Check a workflow for secrets in its parameters before you open its page.
• The Execute Command node is excluded by default from n8n 2.0 (n8n's breaking-changes page, read on 12 November 2026).
• The original copy had no authentication, and its own sticky note says so. Basic Auth does not fit this page: n8n
  renders webhook HTML in a sandboxed iframe, where the headers for Basic Auth are not available. The allowlist is the
  guard here.
• The docsify path and the id of single workflow must match, or the page loads and every file request fails.
• instance_url is built from N8N_PROTOCOL and N8N_HOST. n8n blocks environment variables in expressions only when
  N8N_BLOCK_ENV_ACCESS_IN_NODE is true, and it is false by default.
• Mermaid is a large file. The first load of a page takes a moment.
• A request for a file name the site does not know gets no answer.
• Your workflows' data goes to OpenAI. The pages and the charts load scripts from jsDelivr.