Chat agent that lists which workflows use which credentials

Chat agent that lists which workflows use which credentials

An n8n workflow in two steps. Step one asks your own n8n for all its workflows, keeps each workflow's id, its name and the
credentials its nodes point to, and saves them in a small SQLite database. Step two is a chat. An AI agent turns your
question into a SQL query and answers from what comes back.

What it reads, plainly

It reads credential types, ids and names. Nothing else about a credential is read. The workflow takes the credentials
entry of each node from n8n's list of workflows, and that entry holds a type, an id and a name. It never calls n8n's
credential endpoints, so no secret, password or token is read or saved. A credential's name is free text, though, and can
carry an email address or a client's name.

What is inside

• workflow.json: the workflow as n8n exports it, 9 nodes beside 4 sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

The copy was written for an older Python runner. Mindket made these changes so it runs on n8n's current one:

• Both Python steps read _input and used item.json and .to_py(). The current runner gives code only _items and, for a
  tool, _query. Mindket rewrote both steps to match, and set Save to Database to Python (Native).
• The tool ran any SQL the model wrote. It now answers only a query that starts with select, which is what its own
  description asks for.
• The agent's system message built the workflow link from window.location. n8n's expressions give a window with nothing
  in it, so the message could not be built. The message now holds <YOUR_N8N_ADDRESS>.
• The chat model named no model, so n8n would fill in gpt-5-mini, which OpenAI shuts down on 11 December 2026. Mindket's
  copy names gpt-5.6-terra, the replacement OpenAI gives for it.
• A sticky note said the database lives in memory and is wiped when n8n restarts. The Python step creates a file, so the
  note now says that.
• No credential is inside.

Before you start

• An n8n instance with a Python task runner. Its Python code can import nothing by default, so json and sqlite3 must be
  allowed through the runner's variable N8N_RUNNERS_STDLIB_ALLOW, for example json,sqlite3. For an external runner, n8n's
  docs, read on 12 November 2026, say to set it in the runner's config file as an env override.
• An n8n API key and the API address of your instance, which ends in /api/v1.
• An OpenAI API key.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Allow json and sqlite3 in the Python runner, as above.
3. In n8n, create an API key. Add an n8n API credential with that key and your API address, and select it in n8n.
4. Add an OpenAI credential and select it in OpenAI Chat Model.
5. Open Workflow Credentials Helper Agent and replace <YOUR_N8N_ADDRESS> in its system message with the address of
   your n8n, such as https://n8n.example.com.
6. Press Test workflow on the manual trigger. Save to Database should report how many workflows it stored.
7. Open the chat and ask a question.

How a run goes

Step one runs when you press Test workflow:

1. When clicking "Test workflow" starts the run.
2. n8n lists every workflow the API key can see.
3. Map Workflows & Credentials keeps three fields for each one: workflow_id, workflow_name and credentials, a list
   of the type, id and name of every credential its nodes use.
4. Save to Database makes the table n8n_workflow_credentials if it is missing, then inserts or replaces one row for each
   workflow.

Step two runs when you send a chat message:

1. Chat Trigger starts a run.
2. Workflow Credentials Helper Agent is a Tools Agent, version 1.6, with OpenAI Chat Model and Window Buffer Memory.
3. Query Workflow Credentials Database is its tool, query_workflow_credentials_database. It runs the SELECT the agent
   writes. Its description tells the agent the table and that the credentials column is a JSON string, so a LIKE search
   works best.
4. The agent answers, and gives a link to a workflow when you ask for one.

Change the look

• Which workflows are read: the options of n8n.
• Which fields are kept: the three assignments in Map Workflows & Credentials.
• What the agent knows and how it answers: the system message and the tool's description.
• The model: OpenAI Chat Model.

Good to know

• The database is a snapshot. Press Test workflow again after your workflows change.
• Python gives each run a fresh process, so the steps share one SQLite file that the runner creates in the folder it works
  from. That folder must be writable, and whether the file survives a restart depends on it.
• The agent's answers, which can include credential names and workflow names, go to OpenAI.
• The workflows you can list depend on your API key.
• Each question costs at least one model call, and two when the agent queries the database.
• It runs from n8n's Chat window in the editor. Nothing in the file opens a public chat.
• The JSON holds no credential.