# Myworkingmemory n8n workflow pack

Generated 2026-10-28, revision 4. Create a **blank inactive workflow**, then import one JSON file through n8n's **Import from file** control. In n8n 2.45.5, this control adds nodes to the current canvas and does not apply the file's workflow settings. The exports set `active:false` and include no account credentials, API keys, private run IDs or pinned customer data; keep the destination workflow inactive during setup.

| File | Result |
| --- | --- |
| `01-curated-software-shortlist.json` | Structured software shortlist, source/license links, requirements, evidence limits, CSV and JSON downloads; explicit fixed preview vs paid search page |
| `02-shopify-product-price-export.json` | Product price/availability snapshot, CSV and JSON downloads, partial/limited coverage and resume notes |
| `03-youtube-caption-source-brief.json` | Caption text, source metadata, timestamped excerpts, optional SRT downloads, missing-caption/partial-delivery report |
| `04-email-verification-routing.json` | Ready and hold queues with reasons, two CSV downloads and JSON report; no sending |
| `05-document-metadata-review.json` | Metadata inspection or explicitly requested clean result, inspection hash, findings, verification, warnings and authenticated document-copy download |
| `06-five-tool-ai-agent-starter.json` | Explanation or one explicitly requested, validated, capped Actor run; public schema lookup through MCP Client Tool |

## Setup

1. Create **HTTP Bearer Auth** credentials in n8n using your own Apify token. Select those credentials in each MCP Client node and the optional metadata download HTTP node. The token stays in n8n's credential store; the template has no credential ID and no token URL parameter.
2. Edit the **Configure input** Code node. It contains the Actor's existing input fields plus workflow controls: `runEnabled`, `resumeRunId`, `chargeCapUsd`, and `resultLimit`.
3. Set `runEnabled:true` once the input matches the task you intend to run, then execute manually. Download JSON, CSV or SRT results from the **Build useful output** node's **Binary** tab.
4. For the AI starter, select a customer-owned OpenAI credential in **Customer chat model**. The default explains the five tools. Execution also requires an explicit task and `executionAllowed:true` in **Configure request**. Model usage has its own billing.

The five deterministic templates include a **disabled daily 9 AM schedule**. Before enabling it, open **Workflow settings** and explicitly choose the timezone you want; select **America/Chicago** to match the supplied example. The timezone in the JSON may not be applied by UI import, and the destination's existing timezone, execution timeout and execution-data settings remain in effect. After choosing a repeated input and setting the timezone, clear `resumeRunId`, enable the schedule node and publish deliberately.

## Execution and billing behavior

- A deterministic execution starts at most one Actor. There is no automatic retry on the start node and no automatic pagination. At most 50 result rows are fetched in this starter.
- The start call waits up to 30 seconds, then the same run is checked with a maximum 45-second wait. The Actor itself has a 120-second timeout. A run that is still pending stops with its run ID; set `resumeRunId` to that ID and execute again to retrieve the same run without another paid start.
- `callOptions.maxTotalChargeUsd` applies to pay-per-event charges. It is not a cap on unrelated provider, model or hosting costs. The Actor's own bounded input limits the work; the charge cap does not itself limit the amount of work produced.
- Library preview returns three fixed sample records and requires blank query/category/platform with offset zero. Change `preview:false` for a paid search page; an empty paid page is billable.
- The Shopify snapshot requests availability, not exact inventory quantities. Preserve its `SUMMARY` report because configured limits can mean partial coverage.
- YouTube returns existing captions; it does not transcribe audio or translate. `listOnly:true` yields an `OUTPUT` language report with no transcript dataset rows.
- Email's default `not-an-email` input is a syntax diagnostic. Replace with up to 50 intended addresses. For recurring provider-backed verification, set `useProviderByok:true` and configure the encrypted **HTTP Custom Auth** credential described below. With this flag false, keyless verification uses the finite pilot only if available. A new run of the same list can incur charges again. A ready queue is a verification result, not a sent email or a deliverability guarantee.
- Document inspection is the default. For cleaning, explicitly set `operation:'clean'`, `inspectedSha256` from the inspection, and `confirmContentChanges:true` if the inspection requires content changes. Choose a charge cap for your plan: cleaning is $4 on Free or $0.02 on eligible paid Apify plans, as checked on 2026-10-28. The default $0.05 cap does not authorize a $4 Free-plan clean. The original remains separate from the cleaned output. Review verification, remaining findings and warnings.

The agent can read schemas but cannot start an Actor through its AI tool. A subsequent Code node validates a fixed five-Actor directory and bounded input before a single capped start. Metadata cleaning is deliberately available only in its deterministic template, after inspection.

## Email provider credential

The email workflow offers two mutually exclusive start branches. For ordinary BYOK verification, create an **HTTP Custom Auth** credential in n8n and enter this structure in its credential JSON field, replacing values **inside the credential editor only**:

```json
{"headers":{"Authorization":"Bearer YOUR_APIFY_TOKEN"},"body":{"mv_api_key":"YOUR_MILLIONVERIFIER_KEY"}}
```

Select that credential in **Start BYOK email run** and set `useProviderByok:true` in **Configure input**. n8n merges the secret body field at request time; `mv_api_key` is not present in the template input or export. The node starts the same email Actor through Apify's documented REST API, then rejoins the MCP polling and output pipeline. Use the same Apify account for both credentials. Supplier credits and Apify service charges are separate. The template rejects `mv_api_key` pasted into its Code node.

The dataset-input feature of the underlying Actor is not offered by this 50-row starter; it deliberately sizes verification from the explicit `emails` array.

## Document copies

After an explicit clean request, the workflow builds an authenticated file URL from the returned store ID and record key and downloads the separate file through **Download returned document**, with redirects disabled. Results appear in **Report and document copy**. An inspect or refused-clean result finishes at **Return inspection report**. If `already_clean:true`, the report says no cleaning changes occurred; a returned document is an unchanged copy.

## Compatibility and validation

Node schemas were checked against released n8n **2.45.5**: standalone MCP Client **1.1**, MCP Client Tool **1.4**, AI Agent **3.1**, and OpenAI Chat Model **1.3**. Deterministic schedules use supported Schedule Trigger **1.3**. The Apify endpoint uses **Streamable HTTP**; legacy `/sse` URLs are not used.

Five deterministic workflows completed **six successful live n8n checks**, and **14 offline tests passed**. The live checks covered software preview, a one-product Shopify snapshot, YouTube caption-language listing and transcript/SRT retrieval, invalid-email routing and synthetic-DOCX inspection. See [verification details](verification.md) for the observed outputs, node versions and exact scope.