The SEO content engine your agent can run

List every client in your organization with their ids (id, name, domain, industry). Call this first — most other tools take a client_id. Free, no scopes required.

Use this when someone gives you their website and list_clients has no match for it: it adds that site to Inbounder (or returns the existing one with the same domain) and gives back the client_id every later step needs. Find or create a client (brand/site) in your organization and return its client_id. Matches an existing client by website domain first, then by name — so calling it twice for the same site returns the same client (already_existed: true, matched_on: 'domain'|'name') rather than duplicating. ALWAYS pass domain when you know it. Fails with client_limit_reached when the plan's client limit is hit. Clients cannot be merged or deleted through this API, so prefer list_clients and reuse over creating; set allow_duplicate only after a human confirms a second client for the same site is intended.

List articles in your organization (id, title, status, scores, client) — optionally filtered by client_id and/or status, paginated. Returns metadata only, never article bodies. Use to discover article_ids before get_article, generate_article, score_content, or publish_to_target.

Use this when someone wants to read a draft before deciding whether to send it to their site. Fetch a single article in full: title, status, keywords, scores, published URL, and the complete content (unless include_content=false; the outline is returned while there is no content yet). Use after list_articles when you need the actual text.

List your organization's active integration connections (WordPress, Google Search Console) with their connection ids, site name/URL, and client scope. Use to pick a connection_id before publish_to_target, or to check whether GSC is connected before get_gsc_performance.

Return your organization's remaining credit balance (total plus subscription/referral/purchased breakdown; -1 = unlimited), what each tool costs, and today's daily-quota usage. Use before a multi-step workflow to confirm you can afford every paid call. Exempt from daily quotas (the per-minute rate limit still applies).

Fetch the brand voice profile (tone, ICP, vocabulary, sample patterns) for a specific client. Use before generating any content for that client so the output sounds like them.

Use this when someone wants blog posts that sound like their own website: it reads the site and learns how the brand writes, so the drafts match it. Scrape a client's website text (homepage, /about, blog index) and extract a structured brand VOICE profile with Claude: tone, formality, vocabulary, style guidelines, sentence structure, plus the ideal customer profile (ICP) and example phrases. The url must be on the domain of one of your organization's clients (or a subdomain of it) — any other site is rejected with url_not_owned and nothing is charged; use list_clients to see client domains. If save_to_client_id is provided, the profile is persisted to that client's brand voice (created or updated) and approved, so generate_article and generate_social_posts immediately write in it. If that client already has an APPROVED voice it is returned untouched and nothing is charged — pass overwrite: true only when a human has asked for it to be rebuilt, since re-analysis replaces settings a human may have tuned.

Find target keywords for a topic — a ranked set with search intent and a qualitative difficulty/opportunity read, shaped by your strategy (quick_wins / authority_building / balanced). Use this to plan what to write before create_content_cluster or generate_article. (Difficulty/opportunity are AI estimates; connect a keyword-data provider for hard volume metrics.)

Use this when someone asks you to "turn my website into blog posts" or to plan what to write: it proposes a set of linked blog posts on one topic for their site and saves each one as an outline for them to review before anything is written. Research a topic and design AND persist a topical-authority content cluster for a client: a pillar article plus N interlinked supporting articles, each saved as a real article (status 'outlined') with a title, meta description, target keywords, and outline. Runs the same planner as the app's Research > Cluster flow: keyword research with difficulty/intent/format analysis, then — when the client has an analyzed brand voice — that brand's copywriting framework, highest-commercial-intent-first ordering, and title-length discipline (call analyze_brand_voice first if it has none; without it the plan falls back to a weaker generic one, reported as planner:'basic'). Creates a new project for the client and returns real ids — feed each article_id to generate_article to write the content. Takes ~2-4 minutes; returns an error rather than hanging if it can't finish in time. Never modifies existing projects or articles. Requires client_id (use list_clients).

Use this when someone asks you to "write articles that can rank on Google": it drafts one blog post in the site's own voice for them to read. It does not publish anything, and it cannot promise a ranking. Write a publish-ready SEO article in the client's brand voice for a title/topic + target keyword. Runs the same pipeline as the dashboard's Generate button: AEO structure (quick answer, key takeaways, definitions, FAQ), sourced statistics only, in-body internal links to cluster siblings, the humanization second pass, and a featured image. Returns the article HTML plus AEO + content-quality scores; if an existing article_id is given, the content, meta description, schema data, and hero image are saved back to it (ready to publish_to_target). Expect ~2-3 minutes per article_id call (ad-hoc title+keyword calls are faster but skip humanize/image). The 'write' step of the content engine.

Run the SEO review board over a stored article: five deterministic seats (format conformance, intent contradiction, SERP coverage, answer shape, self-grading) plus one LLM substance seat, then AT MOST ONE bounded redline that rewrites only the sections carrying findings and leaves every other section byte-identical. Records the verdict, the findings and the review timestamp on the article. Free, and a second call against the same body returns the recorded verdict without re-running anything. Findings quote the bytes they rest on, and each carries a grounding tier — a finding grounded in our own format rules is labelled as that, never as 'what ranks'. The 'check' step of the content engine: run it after generate_article and before publish_to_target.

Score a piece of content against a target keyword. Returns AEO (Answer Engine Optimization — readiness for ChatGPT/Perplexity/Google AI citation) and content quality scores. Useful for evaluating articles written outside Inbounder before publishing.

Turn an article into channel-ready social promotion — one tailored post per requested platform (LinkedIn, X/Twitter, Instagram, Facebook, Threads), in the brand's voice. Accepts an existing article_id (content is fetched for you) or raw content. Returns the posts synchronously as drafts — nothing is posted anywhere; a person copies and posts them.

Use this when someone wants to "publish them to WordPress without pasting from a doc": it sends one article straight to their connected WordPress site. The first send creates a draft post that goes live only when they publish it there; sending the same article again overwrites that post. Call it only after the person has read that article and told you to send it. Publish an Inbounder article to a connected destination. Returns the post's URL and target-side ID; a new WordPress post is a draft, so that URL is not a public page until the person publishes it there. Idempotent: republishing an already-published article updates the existing post instead of duplicating it (but still costs 1 credit). Requires write:articles AND publish:wordpress.

Confirm that an article is live at a URL published outside a connected CMS — the manual counterpart to publish_to_target. Records the URL as the article's live location and marks the article published, so Content Health distribution tracking counts it. Idempotent: re-confirming replaces the recorded URL. Inbounder records the URL exactly as confirmed — it does not fetch or verify the page, so only confirm URLs that are actually live. Free.

Use this when someone asks you to "show me which pages Google already sends traffic to", or which searches find their site: it reads their Google Search Console clicks and impressions (group_by "page" for pages, "query" for searches). Return GSC performance for a client over a date range: clicks, impressions, CTR, and average position, aggregated by query, page, country, or device (group_by). Reads from synced data — requires the client to have GSC connected and the daily sync cron to be active.

Use this when someone asks you to "see if ChatGPT and other AI answers mention my site". Tell them plainly what it measures: mention rates come from Claude (web search) and Gemini (Google-grounded) answers; ChatGPT answers are not checked yet. Report AI-search (GEO) visibility for a client or the whole org. Returns TWO separate signals: (1) readiness — how well each article is optimized to be cited by AI answer engines, based on Inbounder's per-article AEO score, with an average, well-optimized/needs-work counts, and a ranked per-article breakdown; (2) measured — actual probe results from Claude (web search) and Gemini (Google-grounded): mention rate WITH sample size (e.g. mentioned in 14 of 42 runs), per-engine splits, top cited source domains, gap domains cited where the brand wasn't, and a third-party source breakdown — what share of citations came from surfaces the brand doesn't own (e.g. review sites, Reddit, editorial), with the top non-owned domains and example URLs. The third-party share is null (never 0%) until citations exist, and is only meaningful for clients with a domain on file — clients_missing_domain counts clients where it isn't, since every citation for those falls into third_party by construction. Measured is null with an explanatory note until probes have run. ChatGPT/Perplexity/Google AI Overviews are not currently probed. Omit client_id for an org-wide report.

Per-dimension Content Health for one client: Production, Distribution, Keywords, and Maturity — each returned as a state (on_track / attention / too_early / no_data) with a 'why' sentence carrying the real counts it derives from, plus the cited reference bands (source, year, method, confidence) the states lean on. This tool deliberately returns no composite health score: a single blended number would hide which dimension needs work and imply measurement precision the underlying research does not support. Free.

Import a client's sitemap and check which of its pages are live. Reads the client's domain: robots.txt Sitemap lines, then /sitemap.xml, then /sitemap_index.xml (sitemap indexes followed one level). Each listed URL is fetched once with a GET; only a 2xx response after redirects on the client's own domain counts as live — 404s, 5xx, timeouts and off-domain redirects are stored as not verified. Capped at 5,000 URLs per client: `truncated: true` means the sitemap listed more and only the first 5,000 were kept. Large sites finish in the background (a scheduled job continues every 15 minutes and re-syncs weekly); `in_progress` and `remaining` say how far it got. Repeating an identical call within ~15 minutes returns the first result, so wait before checking progress. This only builds the list of live pages — it does not mark articles as published, and get_content_health's live counts do not change from it yet. Requires the client to have a domain. Free.