adscapi

One canonical conversion event, fanned out to every ad platform's server-side Conversions API and audience-sync API. Fire one typed event; adscapi hashes the identity, maps the event name per platform, and dispatches to each one you've configured.

Ships four ways from one package: an SDK you import, a CLI, an MCP server any agent can drive, and a hosted relay so a backend in any language can POST one event over HTTP.

import { createAdscapi } from 'adscapi';

const ads = createAdscapi(); // reads platform tokens from the environment

const results = await ads.conversions.track({
  name: 'purchase',
  value: 49,
  currency: 'USD',
  user: { email: '[email protected]' },   // raw — adscapi hashes advanced-matching fields per platform
  clickIds: { fbclid: '…', gclid: '…' }, // captured server-side
  consent: { adUserData: true, adPersonalization: true }, // optional: gppString, tcfString, limitedDataUse
});

// results: [{ platform: 'meta', ok: true }, { platform: 'tiktok', ok: false, error: '…' }, …]

A platform is active only when its secrets are present, so the same call works whether you run one platform or twenty-six. One platform failing never blocks the rest — you get a result per destination instead of an exception. user accepts the full advanced-matching set (name, city, state, zip, country, DOB, gender, externalId, ip, userAgent, fbc/fbp); more match keys raise match rate and lower CAC. consent also accepts optional IAB gppString / gppSectionIds / tcfString and a limitedDataUse flag (CCPA). GA4 always gets Consent Mode (GRANTED/DENIED) so Google can model conversions under signal loss; Meta gets Limited Data Use when the flag is set.

Why

Every ad platform wants the same thing — "this user converted, here's the hashed identity" — but each has its own Conversions API, its own event names, its own token dance. The official SDKs are one platform each; the hosted tools (Servo, Stape) are no-code SaaS; the CDPs (Segment, RudderStack) are heavyweight and ad-tracking is a side feature. adscapi is the missing middle: a code-first library that does ad-conversion fan-out and nothing else.

What you get

• Conversions — ads.conversions.track() fans one event to 26 platforms' server-side Conversions APIs. PII hashing, per-platform event mapping, cross-platform dedup, retries, dryRun, and a per-call destinations toggle are built in. track(event, { testEventCodes: { meta: 'TESTxxxxx' } }) sends a real event to the platform's Test Events view (vs dryRun, which sends nothing).
• Audiences — ads.audiences.sync() uploads a hashed customer list to 28 platforms' audience APIs.
• Offline imports — uploadOffline() batch-uploads offline conversions from a CSV to Bing, Google Ads, and Meta.
• Creatives — uploadCreative() uploads an ad image (Meta today; more landing per PR).
• Agent commerce — fromAcpCheckoutSession() / fromAp2PaymentMandate() turn a settled ACP or AP2 agent purchase (no browser, no pixel) into a canonical event you track. So the ad that drove the agent still gets credit. See docs/AGENT-COMMERCE.md.
• Measurement read-back — getReport() pulls spend, impressions, conversions, and ROAS back from Meta, Google Ads, and TikTok to close the loop. See docs/REPORTING.md.
• Governance — Governor gates dispatch with a kill-switch, approval gate, rate cap, and audit trail for agents running under governed autonomy. See docs/GOVERNANCE.md.
• Observability — a queryable, restart-safe delivery log via the onResult hook. See docs/OBSERVABILITY.md.
• AdCP interop — map results onto the AdCP (metric_id, qualifier) measurement taxonomy. See docs/ADCP.md.
• Four faces — the SDK above, a CLI, an MCP server for agents, and a hosted relay for non-JavaScript backends.

Run it from any language or runtime

• Import the SDK on any modern JS runtime — Node 18+, Bun, Deno, Next.js, and edge runtimes like Cloudflare Workers. The dispatch path uses only fetch and WebCrypto, no Node built-ins.
• Any other language — run the hosted relay (a Cloudflare Worker) and POST one event to it over HTTP from Python, Ruby, Go, PHP, Rust, anything. Or shell out to the CLI from CI.
• Agents — point any MCP client at npx adscapi mcp.

adscapi ships no per-framework SDK — there's one library, and you call it from your stack. The relay and CLI cover everything that isn't JavaScript.

CLI

npx adscapi platforms      # every destination + the secrets it needs
npx adscapi init           # interactive setup: prompts for missing secrets, writes .env
npx adscapi check          # which platforms are configured right now
npx adscapi verify         # live-check that configured tokens actually work
npx adscapi doctor         # health check across configured platforms (nonzero exit on failure)
npx adscapi test-event --event purchase --email [email protected] --dry-run
npx adscapi offline --platform bing --file conversions.csv
npx adscapi creatives upload --platform meta --image https://cdn.example.com/hero.jpg

doctor exits non-zero when any configured platform fails its live check, so you can run it in cron or CI as a health gate.

check reads your environment; verify goes further and calls each platform's API to confirm the token is valid — so you catch an expired token before a real event does.

Hosted relay (non-JavaScript backends)

Deploy the Cloudflare Worker in worker/, set your platform secrets on it, and any backend can fan out with one HTTP call:

curl -X POST https://<your-worker>/track -H "Authorization: Bearer $TOKEN" \
  -H 'Content-Type: application/json' \
  -d '{"name":"purchase","value":49,"currency":"USD","user":{"email":"[email protected]"},"consent":{"adUserData":true,"adPersonalization":true}}'

Platform secrets live on the Worker, never in the caller's request. See worker/README.md.

MCP

// .mcp.json
{ "mcpServers": { "adscapi": { "command": "npx", "args": ["-y", "adscapi", "mcp"] } } }

Lowering your CAC

Fan-out is table stakes; match quality is where cost per acquisition actually moves. docs/MATCH-QUALITY.md is the playbook: which identifiers to send, how to optimize for value, how to dedup and protect the signal under consent loss, and how to read Meta's match-quality score as you tune.

Per-platform setup

Each destination has a setup guide in docs/platforms/: how to create the account, where each token comes from, and how to verify it. Start with Meta. Each conversion platform records the API doc it was built against; npm run check:docs reports which are freshly verified and flags any dead doc links.

Status

Live on npm. Twenty-six conversion platforms and twenty-eight audience-sync platforms ship against their real API docs, plus offline CSV import (Bing, Google, Meta) and image creatives (Meta). Five conversion platforms carry a live credential check today; the rest are marked unverified and re-checked over time. The agent-native suite — ACP/AP2 checkout adapters, spend/ROAS read-back, governance, a delivery log, and AdCP measurement interop — ships alongside. Branch model: main is staging, release publishes to npm.

License

MIT