MCP and connecting your AI

If you already do strategy and implementation in Claude Code, Codex, or another MCP client, you should not have to export half the product into a paste. EZ SEOS exposes an MCP server so that client can call the same project tools the in-app agent uses: research, ranks, Visibility, audits, analytics, and more. The catalog is 60+ tools.

You connect with OAuth consent, optionally pin a project-scoped connection URL so client data never mixes, install or edit SEO skills, and lean on project memory that stays scoped to that project. How the agent should think and when a human has to approve is the agent (/docs/the-agent). Commercial angle for the agent plus MCP: SEO agent (/features/seo-agent).

Why connect your own AI?

Strategy work already lives in the model you use to implement. MCP keeps that work where you already are, with live tools on the project instead of a second chat full of screenshots.

Grounded behavior still applies. Paid actions show cost. When tools are used correctly, the model should not invent metrics the tools did not return.

What you need

• An EZ SEOS hosted account with access to the project you care about (private testing: access is gated, not an open free-for-all on the public internet)
• An MCP-capable client (Claude Code, Codex, or another client that speaks MCP)
• Permission to authorize the connection through OAuth

If you manage several client sites, plan on project-scoped URLs so each connection has a clear home. See projects (/docs/projects).

OAuth consent

Connection uses a normal login and consent flow. External agents only get the org and project data you authorize. You stay in control of that grant. Docs stay at the consent-concept level. You do not need to paste long-lived secrets into a setup page to understand how this works.

Project-scoped connection URLs

You can pin MCP to one project so the agent does not wander across clients. For agencies, one connection per client project is the clean pattern. See projects (/docs/projects). Broad URLs that touch every project are how people mix data by accident. Pin the project.

The tool catalog (60+ tools)

The product exposes a live catalog of registered tools, grouped by category, including which ones spend credits. Expect groups in the neighborhood of:

• Keywords, SERP, ranks
• Domain intelligence / Search Truth
• Backlinks / prospects
• Audits / originality / page content
• Visibility Engine / Do Next / GEO actions
• CTR, local, analytics, GSC, IndexNow
• Page Glance, reports/export, digests

Open the catalog in the product for the current list. It will move as tools ship. This page is not a frozen dump of internal function names, and it does not document server paths, hosting regions, or provider names.

Installable and editable skills

Skills are how you make the external (or in-app) agent work the way your team works. You get pre-seeded SEO skills you can edit, enable or disable, and reset. Packaged skills act like installable playbooks for common workflows: project setup, keyword research, clustering, competitive landscape, link prospecting, content-gap work, GEO site changes, CTR autopsy, orphan rescue, ship-and-verify, and similar.

Treat them as editable playbooks, not a tour of a code repo. Turn on what your process needs. Turn off what gets in the way.

Project memory

Project memory holds durable facts across sessions. You can edit, delete, or forget. It is scoped to the project and does not cross clients. Whether you chat in-app or through an MCP-connected client, you work against that same project-scoped memory model.

Credits and approvals when tools run

MCP tools that spend credits still follow product cost discipline: paid scans and pulls show cost, and human approval remains for paid work that needs it. Free tools stay free (including GSC and GA reads and similar free surfaces). Credits overview: credits (/docs/credits).

How this relates to the in-app agent

Two surfaces, same underlying capabilities.

• In-app agent: judgment, conversation inside EZ SEOS, docs on the agent (/docs/the-agent)
• External MCP: your preferred client, same project tools, this page

Pick the surface that matches how you already work. Do not treat MCP as a second product with a second set of made-up metrics.

Getting access

Hosted access is in private testing. Soft path: join the waitlist (/waitlist) or early access flow (/docs/early-access) when you are ready. When you have a project, connect from an MCP client with OAuth and a project-scoped URL.

Can I use ChatGPT desktop?

Use any MCP client that supports the protocol. Do not assume every chat UI ships MCP support. If the client cannot speak MCP, it cannot use this connection.

Is data mixed across clients?

Not if you pin project-scoped connection URLs and keep one connection per client project. Mixing happens when people share broad access on purpose or by accident.

Do I need to paste API keys into every chat?

No. OAuth consent is the authorization model described here. Stay in the product consent flow rather than scattering long-lived secrets in prompts.