• Nov 2, 2026
  • 5 min read

Every Integration Behind a One-Person Lead Engine: The Map

Over the last few months this site has turned into a build log: an inbound lead engine, an outbound pipeline, and the weekly habit of reading what both return. Each article covers one piece. This one is the map: every external service and API in the system, what data it moves, and where the gaps are.

The whole picture

flowchart TB
    subgraph OUT["Outbound"]
        direction LR
        AP[Apify<br/>maps listings] --> EN[Own-website fetch<br/>+ Companies House API]
        EN --> AU[Audit engine<br/>Lighthouse]
        AU --> CSV[send-ready CSV]
        CSV --> ST[Sending tool<br/>CSV in, CSV out]
        ST --> LK[(DuckDB funnel)]
    end
    subgraph IN["Inbound"]
        direction LR
        FM[Forms on every site] --> WK[Cloudflare Worker + KV]
        CAL[Cal.com webhook] --> WK
        WK --> MJ[Mailjet<br/>+ event webhook]
        WK --> NT[ntfy alerts]
        WK --> HL[(Homelab SQLite<br/>pull-only mirror)]
    end
    subgraph SEE["Visibility"]
        direction LR
        CWA[Cloudflare Web Analytics] --> REP[make analytics]
        GSC[Search Console API] --> REP
        PSI[PageSpeed Insights API]
        BW[Bing Webmaster Tools]
    end
    OUT ~~~ IN ~~~ SEE

Outbound: from a listing to a reply

IntegrationWhat it doesAccessDetail
Apify (Google Maps actor)Business listings: name, website, address, category. Reviews, photos and people switched offAPI tokenFrom Google Maps to a UK lead list
The business’s own websiteThe email address it publishes, including ones Cloudflare obfuscatesPlain HTTP, own User-AgentFinding a business email on its own website
Companies House APIActive limited company? Registered name, number, officeFree API keyMatching without false positives
Audit engine (Lighthouse, run locally)One measured problem per site, measured twiceNone (local)Personalised with an audit, Lighthouse variance
Sending toolWarmed inboxes, sequence, A/B subject, blocklistNo API on my plan: CSV import, CSV exportSending capacity
DuckDB (local file)Bronze files → silver tables → gold countsNone (local)The Sunday review

The why and the rules behind it: is B2B cold email legal?, keep outbound away from inbound, the pipeline end to end.

Inbound: from a form to a follow-up

IntegrationWhat it doesAccessDetail
Cloudflare Workers + KVOne forms endpoint for every site, same-origin; retention as a TTLwrangler login / API tokenOne forms Worker, retention as a KV TTL
MailjetConfirmation and drip emails; an event webhook for bouncesAPI key + secret; webhook keyValidating a sending subdomain, the event webhook
Mailtrap / MailpitA second provider behind the same adapter; local test captureAPI token / noneOne sender adapter, Mailpit vs Mailtrap
Cal.comBookings into the same pipelineSigned webhookCal.com signed webhook
ntfy (self-hosted)A push to my phone per lead; a weekly counts-only digestWrite-only publisher token, behind a tunnelSelf-hosting ntfy, weekly digest
Homelab mirrorKV copied to SQLite, pulled from home, no inbound portRead-only Cloudflare API tokenA pull-only homelab
Pages deploy hooksNightly rebuild so future-dated posts appearHook URL (a secret)Future-dated posts

All of it, in order: the inbound series.

Visibility: is anyone finding the sites?

IntegrationAnswersAccessDetail
Cloudflare Web AnalyticsVisits, pages, referrers, countries, browsers; no cookiesGraphQL with the wrangler loginThe GraphQL API
Google Search ConsoleQueries, clicks, impressions, position, sitemapsService account JSONStep by step, API key vs service account
PageSpeed InsightsLab speed from Google’s machines, real-user data when there is anyAPI key (public data only)API key vs service account, Lighthouse variance
Bing Webmaster ToolsBing’s index and trafficVerification (DNS, meta tag or file)Three ways to verify

The first two sit behind one command: a monthly website check. What it led to: separate SEO titles.

Five kinds of credential

Seventeen integrations, but only five ways of proving who I am:

KindUsed forRule
API token / keyApify, Companies House, Mailjet, PageSpeed, ntfy, the homelab mirrorIn a config folder outside every repo; scoped to the minimum
OAuth loginCloudflare via wranglerShort-lived; refresh before scripted use
Service account JSONSearch ConsoleNo project roles; added only where it’s needed, read-only scope for reports
Signed webhooksCal.com, Mailjet eventsVerify the signature before reading the body
Public by designWeb Analytics beacon token, Bing verification codeFine in the HTML; still not something to reuse elsewhere

Where the data still doesn’t meet

Closing the gaps is the point of most of these articles. These are the ones still open, on purpose:

  • The sending tool has no API on my plan. Results come out as CSV every Sunday, by hand. Five minutes a week is cheaper than the next plan up.
  • Outbound and inbound don’t share a database, by design: different legal basis, different retention. They only meet as counts in a review.
  • Bing isn’t in the monthly report yet. It will be if it keeps sending more visitors than Google.
  • Outbound has no click data: the emails carry no links and no tracking. The only signal besides a reply is a visit to the outbound brand’s own site, which now has the same cookieless beacon.

Takeaways

  • Write down every integration once: what it moves, how it authenticates, and where it’s explained.
  • Few credential types, kept outside the repos, beat many clever ones.
  • Not every gap needs closing. A weekly CSV is an integration too.