Skip to content
API

An AI SEO API for agents — MCP today, REST coming soon

An AI SEO API gives your own tools programmatic access to AI-search data: whether ChatGPT, Perplexity, Gemini and Claude cite and recommend your brand, your share of voice against competitors, the real queries the models ran, and the verified AI-crawler and AI-referral traffic hitting your site. SourceWatch ships that access in the form agents actually use first — a Model Context Protocol (MCP) server that runs **today** inside Claude Code and any MCP-compatible client. A language-agnostic public **REST read API is coming soon** (waitlist via contact), with **webhooks planned** after it. This page is honest about that order: MCP is live, REST is roadmap.

TL;DR

  • **Live today: the MCP server.** Query your AI-visibility and citation data directly from your AI agent — Claude Code or any MCP client — over an open standard. No REST plumbing required.
  • **Coming soon: a public REST read API.** Language-agnostic HTTP endpoints for visibility, citations and prompt data that fit your existing CI and dashboards. **Waitlist only** — join via contact. Do not expect a documented REST endpoint today.
  • **Planned: webhooks.** Push notifications on citation and visibility changes, so you stop polling. Explicitly on the roadmap, not shipped.
  • **Why MCP first is credible, not a workaround:** MCP is an open standard adopted across the industry in 2025 — OpenAI in March, Google for Gemini in April. For agent workflows, an MCP server *is* the native API.
  • **The data behind the API — two moats:** first-party AI traffic **verified against each vendor's published IP ranges** (ground truth, not a synthetic sample), and an MCP surface you can **act on in-loop**, not just read.
  • **Honest scope:** SourceWatch returns measurement and briefs, **not generated content**; the **REST API is not live yet** (MCP only); the **free audit is one page** — full-site is the 14-day trial (card optional).

Where the AI SEO API stands today

Most "AI SEO API" pages quietly imply a REST endpoint that either doesn't exist or sits behind a sales call. SourceWatch is built the other way around: the access that works right now is the MCP server, and the REST API is on a published roadmap you can hold us to. Here is exactly what is shipped, what is next, and what is planned — in plain terms.

PhaseWhat it isStatusHow you get it
v0 — MCPQuery visibility, citations & prompt data from your AI agent over the Model Context ProtocolLive nowConnect the MCP server in Claude Code / any MCP client
v1 — REST read APILanguage-agnostic HTTP read endpoints for visibility, citations & prompts — fits CI & dashboardsComing soonJoin the waitlist via /contact
v2 — WebhooksPush notifications on citation / visibility changes so you don't pollPlannedRoadmap — follows the REST API

Read this before you build against it

If your integration **needs a documented REST endpoint today**, the honest answer is: not yet — that's exactly why it's on a waitlist, not a pricing page. If your integration runs **through an AI agent** (and increasingly they do), the MCP server is live and is the faster path anyway. Pick the row above that matches your stack.

For agents, the MCP server is the API

It's tempting to read "MCP, not REST" as a compromise. It isn't. The Model Context Protocol is an open standard, launched by Anthropic in November 2024 as a way to build secure, two-way connections between data sources and AI-powered tools. It replaces a pile of bespoke one-off integrations with a single protocol — the same role REST played for the last era of the web, but built for agents.

And it isn't a single-vendor bet. Through 2025 the rest of the industry converged on it:

  • **OpenAI adopted MCP** in March 2025, integrating it across products including the ChatGPT desktop app.
  • **Google DeepMind confirmed Gemini support** in April 2025.
  • **Microsoft and major coding platforms** (including Replit and Sourcegraph) followed, wiring MCP into developer tooling.
  • The protocol now ships official SDKs in Python, TypeScript, C#, Java and more — so an MCP server is a first-class integration target, not a niche one.

Why this matters for an AI SEO API specifically

Your buyers ask AI engines the questions; your team increasingly works *through* AI agents too. Shipping an MCP server first isn't skipping the API — it's putting the API where the work already happens. Your assistant reads your SourceWatch citation and traffic data the same way it reads your repo or your docs: natively, over a protocol it already speaks.

There's a second, quieter advantage. A REST client *reads* data and then a human decides what to do. An MCP client reads the same data and can **act on it in the same loop** — pull your citation gaps and the real queries the models ran, then draft answer-first content against them in Claude Code without leaving the editor. That act-in-the-loop workflow is the second moat, and almost no legacy SEO tool offers it; where a comparable agent layer exists, it's enterprise-only and gated behind a separate subscription. SourceWatch puts it on a self-serve plan.

What the API exposes (today via MCP)

Whether you reach it through the MCP server now or the REST read API later, the underlying data is the same — the two sides of AI search SourceWatch measures. Today, from inside an MCP client, your agent can read:

Visibility & citations

  • **Mention & self-citation rate** — how often ChatGPT, Perplexity, Gemini and Claude name you, and whether they cite your own pages.
  • **Share of voice** — your slice of brand mentions versus the competitors the models name instead, on a fixed prompt set, over time.
  • **The real queries the models ran** — the actual prompts behind each answer, plus sentiment, most-cited domains, and the citation gaps to close.

First-party AI traffic

  • **AI-crawler analytics** — which AI bots (GPTBot, ClaudeBot, PerplexityBot, Google-Extended and more) read which pages, how often — see AI crawlers.
  • **AI-referral visitors** — the real people who arrived from an AI answer, captured first-party instead of lost to "direct." See AI traffic analytics.
  • **Verification metadata** — every crawler and visitor checked against the vendor's published IP ranges, so spoofed user-agents don't pollute what your agent reads.

The data quality is the point of an API

An API is only as good as the numbers it returns. SourceWatch's first-party capture is **verified against published vendor IP ranges**, not guessed from user-agent strings — the difference between "something claimed to be GPTBot" and "this was verifiably OpenAI." Because it comes from your own logs rather than a synthetic prompt sample, what your agent pulls is ground truth, not an inference that drifts with prompt selection.

The REST API: what's coming, and what to expect

The public REST read API is the next phase, and it exists for the integrations MCP doesn't naturally cover: a language-agnostic HTTP surface that drops into an existing CI pipeline, a BI tool, or a customer-facing dashboard — anywhere a plain authenticated GET is the right shape. It will be token-authed and account-scoped, the same per-account isolation model the platform already runs on.

What to expect, stated plainly:

  • **It is read-first.** Visibility, citations and prompt data out; not a write API.
  • **It is not live yet.** Access today is the MCP server. There is no documented REST endpoint to point at right now — and we won't pretend there is.
  • **Webhooks follow it.** Push on citation/visibility changes is planned *after* the read API, so you can stop polling once both land.
  • **You join a waitlist, not a checkout.** Tell us your stack via contact and we'll bring you in as REST opens up.

Need a language-agnostic REST endpoint for CI or a dashboard? Join the REST API waitlist and tell us your use case — we'll bring you in as it opens.

Join the REST API waitlist

Why an AI SEO API matters now

Programmatic access only matters if the channel it measures is real — and AI search has crossed that line. ChatGPT reached more than 800 million weekly active users, announced by OpenAI's Sam Altman at its October 2025 Dev Day — up from 500 million at the end of March 2025. (Source: OpenAI, via TechCrunch.) When discovery moves into AI answers at that scale, "are we cited?" becomes a number you want flowing into your own systems, not trapped in someone's dashboard.

800M+

weekly active ChatGPT users, announced by OpenAI at its Oct 2025 Dev Day — up from 500M at the end of March 2025 (third-party reporting, TechCrunch)

The referral channel is climbing just as fast. Third-party measurements over autumn 2025 put ChatGPT referrals up roughly 52% year over year and Gemini referral traffic up around 388% in the same window (Similarweb data, via Digiday) — directional estimates, not our figures, but they all point the same way: AI is becoming a high-intent acquisition channel, and one most analytics stacks still misattribute. An API is how you wire that signal into the tools your team already runs.

Grounded in published research, not vibes

The optimization model SourceWatch measures maps to the peer-reviewed GEO (Generative Engine Optimization) paper, accepted at KDD 2024, which showed that making content quotable — adding citations, quotations and statistics — can lift visibility in generative answers by up to ~40%, benchmarked across the paper's GEO-bench query suite. The same work found classic SEO tricks don't transfer: tactics like keyword stuffing failed to help and often hurt. So the API returns metrics tied to a real method, not a made-up score.

Who it's for

The AI SEO API is for the people who want the data in their pipeline, not just on a screen:

  • **Technical teams already in Claude Code** — wire SourceWatch in via MCP today and run the measure-and-act loop without a separate dashboard.
  • **Agencies building client reporting** — pull per-client visibility and AI-referral data into white-label reports; the REST API (on the waitlist) is the clean fit for automated, scheduled exports.
  • **In-house data & RevOps** — land AI-citation and share-of-voice metrics in your BI tool or warehouse alongside the rest of your acquisition data.
  • **Builders & platforms** — anyone who wants AI-visibility data as a feed for their own product, once the REST surface opens.

If you're evaluating the broader product rather than just the integration, start at the AI visibility tracker hub or browse the best AI SEO tools roundup — the API is how you reach the same data programmatically. New to this? See how it works and the full feature set.

Get started

Two honest paths, depending on what you need:

  1. 1

    Use MCP today

    Start a 14-day trial (card optional), connect a site, and add the SourceWatch MCP server in Claude Code or your MCP client. Your agent can read your visibility, citation and traffic data — and act on the gaps — within about 15 minutes of connecting.

  2. 2

    Waitlist REST for later

    Need a language-agnostic HTTP endpoint for CI, a warehouse, or a dashboard? Join the REST API waitlist via /contact and tell us your stack. We'll bring you in as the read API opens; webhooks follow.

  3. 3

    Check your starting point first

    Not sure you're even visible yet? Run the free single-page AI SEO audit — it checks whether AI engines can read and recognize one URL in about 15 seconds, no card required. A full-site scan runs on the trial.

Get the live MCP server and your AI-visibility data inside Claude Code today — REST and webhooks are on the way.

Start your free trial

Frequently asked questions

Does SourceWatch have an AI SEO API?

Yes — today via an MCP server, with a public REST API coming soon. The MCP (Model Context Protocol) server is live now and works natively inside Claude Code and any MCP-compatible client: your AI agent can read your AI-visibility, citation, share-of-voice and AI-traffic data, and act on it in the same loop. A language-agnostic public REST read API is on the roadmap (waitlist via /contact), with webhooks planned after it. We're deliberate about that order: MCP is shipped, REST is not yet.

Is there a public REST API available right now?

Not yet — and we won't pretend otherwise. Today's programmatic surface is the MCP server. The public REST read API is "coming soon" and access is via a waitlist on the contact page, not a self-serve checkout. If your integration hard-requires a documented REST endpoint this minute, that's a fair reason to wait or talk to us about timing. If it runs through an AI agent, the MCP server is live and is the faster path anyway.

Why ship MCP before REST?

Because for agent workflows, the MCP server is the native API. The Model Context Protocol is an open standard launched by Anthropic in November 2024 and adopted across the industry in 2025 — OpenAI in March and Google for Gemini in April, with Microsoft and major coding platforms following. Shipping MCP first means meeting AI tools where the standard already converged, rather than building a bespoke REST surface they'd have to wrap. A REST API still follows, for the language-agnostic, non-agent integrations MCP doesn't naturally cover.

Source: Anthropic — Introducing the Model Context Protocol
What data can I pull through the API?

The two sides of AI search SourceWatch measures. On visibility: mention and self-citation rate across ChatGPT, Perplexity, Gemini and Claude, share of voice versus competitors, sentiment, the real queries the models ran, most-cited domains and citation gaps. On traffic: which AI crawlers read which pages, and the real AI-referral visitors who arrived — all verified against each vendor's published IP ranges, so the data is ground truth rather than guessed from user-agent strings.

When will the REST API and webhooks ship?

We're not putting a hard date on it, because we'd rather under-promise here than miss. The order is fixed: the REST read API comes first (it's in active roadmap, waitlist open via /contact), then webhooks for push notifications on citation and visibility changes. Join the waitlist and tell us your use case — that genuinely influences sequencing, and we'll bring you in as each phase opens.

Can the API generate content or fix my pages?

No. SourceWatch returns measurement and content briefs — your citation gaps, the real queries, and what to write — but it does not generate finished content or auto-publish. Because it's MCP-native, your assistant in Claude Code can pull that guidance and draft against it in the same loop, but you (or your team) own the writing. We're upfront about this so you can choose with eyes open.

How is the API authenticated and scoped?

Access is per-account and token-authed. The MCP server connects against your account today; the forthcoming REST API will use the same per-account isolation and token model, so data stays scoped to your own sites. There are no extra seat costs to integrate — every plan includes unlimited seats, and a 14-day free trial with card optional. See pricing for plan details.

Is the AI SEO API enough on its own, or do I need the dashboard?

You can run entirely through the MCP server if your workflow lives in Claude Code — read your visibility and traffic data and act on gaps without opening the app. Most teams use both: the AI visibility tracker dashboard for trends, alerts and share-of-voice charts, plus the API to pull the same data into their own pipeline. The data is identical on both surfaces; the API just removes the screen between you and the numbers.

Source: GEO: Generative Engine Optimization — Aggarwal et al. (KDD 2024)

Further reading

Keep reading

Live today: query your AI-visibility, citation & AI-traffic data from your agent over MCP. REST read API and webhooks coming soon — waitlist via contact.

Connect your first site and watch SourceWatch score your AI visibility in minutes.