For Developers

A portable autonomous SEO engine — six ways to ship.

Pure ESM cores, a two-seam adapter contract, and a trusted/untrusted split that keeps provider keys server-side. Architecture, integration surfaces, the operational contract and a safe adoption path.

Prepared for I-Novanet Software Ltd / AscentUpdated 23 Jul 2026

Value proposition. Embed search intelligence without rebuilding research, prioritisation, generation, publishing, verification and governance. The same core powers an SDK call, HTTP service, MCP tool, assistant skill, browser widget or operator cockpit — the application chooses its storage and content adapters.

Verified runtime profile

Property	Statement
Module system	Pure ESM; no build step
Dependencies	Zero runtime dependencies
Runtimes	Node, Deno, edge and browser
Verification	`node --test`; 560+ tests
Core facade	`createSeoAgent()`
Integration seams	Content and storage adapters

Six delivery surfaces

Surface	Use when	Operational boundary
SDK	Embedding inside a trusted service	Facade over the cores; secrets server-side
HTTP service	Language-neutral / centralised	Request→Response router; bearer-licensed
MCP server	Giving AI clients SEO tools	Zero-dep stdio server; six SEO tools
Claude skill	Assistant-native workflow	MCP-preferred with script fallback
Browser widgets	Showing ranks, potential, results	Pure build/mount presentation components
Admin cockpit	Operating proposals, keywords, rankings	Host controls persistence through onAction

The action lifecycle

Sense external and first-party signals.
Create a typed proposal with rationale, confidence, expected value and cost.
Apply policy: off, human-approved or automatic for the task type.
Execute through the host's content adapter and record the action.
Verify the live result and measured rank/traffic outcome. measured
Revert regressions, trip circuit breakers, feed the learning record.

const agent = createSeoAgent({ storage, content, providers, budget, policy })
const plan = await agent.run({ site, goal })
await host.reviewOrApply(plan.actions)

Conceptual pseudocode — confirm exact package paths in the portal/download bundle.

Safe adoption pattern

Stage	Policy	Exit criterion
Observe	All task types off; collect proposals + costs	Stable signal collection, trusted baselines
Approve	Human review for every action	Repeated acceptance + safe verification by task type
Graduate	Auto only for high-confidence, reversible tasks	Low regression rate within spend + brand thresholds
Operate	Scheduled loop with alerts + periodic review	SLOs, audit retention, rollback rehearsed

Production checklist

Verify domain ownership and tenant boundaries before write access.
Keep search, LLM, CMS and analytics credentials in the trusted service.
Idempotency, quotas, timeouts, retry limits and circuit breakers for every provider.
Separate measured Google rankings from Ascent estimates in schemas and UI.
Retain immutable action, approval, publish, verification and rollback events.
Test adapters against staging; restrict automatic tasks to reversible changes first.
Define data retention, regional processing, incident response and dependency ownership before enterprise launch.

Current boundary & roadmap

Available now

SDK, service, MCP, skill, widgets, admin cockpit; scheduled runs; publishing; self-healing; budget controls; multilingual; GEO readiness; and measured AI-answer citations (Perplexity). Card billing (MIPS) + admin back-office live; free beta open.

Next commercial layer

Marketplace identity & billing (I-Novanet Software Ltd); broader measured AI-citation coverage across engines; pooled cross-site learning (foundation shipped; model pending data); broader packaged connectors and enterprise controls.

Use the SDK for in-process control, the HTTP service for organisational reuse, MCP/skills for agent ecosystems, widgets for embedded evidence, and the cockpit for human governance.

Open the developer portal →