Seonix

npx add-mcp 'https://api.seonix.ai/mcp'

Your SEO data inside the agents you already use

The Seonix MCP server lets the AI agent you already work with read your Site Health report and fix plan, pull Search Console queries, browse keyword clusters and articles, and score a draft before it goes live. Ask in plain language; the agent calls Seonix.

>Connect Seonix to this project over MCP: https://seonix.ai/mcp.md

Paste this into Claude Code, Cursor, Codex or any agent with MCP support. It reads the manifest, adds the server and opens the sign-in; you approve once in the browser.

Sign in with OAuth: approve once in the browser, revoke any time. Every connected agent is listed in your account. Part of every paid plan. app.seonix.ai/account/agents

Sign-in is OAuth — approve once at app.seonix.ai, revoke any time under Account → Agents. API keys remain for scripts. The plugin also ships three skills — seonix-seo, seonix-fix-site and seonix-content — so the agent knows the Seonix workflows without reading the manifest. Plugin source on GitHub

See what the agent can do

Example session · Claude Code

U

Fix everything Seonix found on my site and tell me what changed.

• get_site_healthSite Health 61 · 43 open problems
• get_fix_plan14 items · 3 site-wide, 11 page-level
• analyze_seorewritten page scores 84/100 A

Done: canonical tags, 9 meta descriptions, alt text on 31 images and the sitemap path in robots.txt. Start a rescan in Seonix and I will confirm what cleared.

After rescan: 61 → 88

Claude Code

Anthropic

Plugin from the marketplace, or one command.

Cursor

Anysphere

One-click button or ~/.cursor/mcp.json.

Codex

OpenAI

Plugin from the marketplace.

Copilot

GitHub

VS Code reads.vscode/mcp.json.

Windsurf

Windsurf

Same mcpServers entry, URL only.

Works with any client that speaks the Model Context Protocol over HTTP: add the URL and sign in with OAuth, or send an API key as a bearer header for unattended runs. Coming to the Claude and ChatGPT directories.

What it is

The product you pay for, reachable from a chat

MCP (Model Context Protocol) is the open standard AI agents use to call outside tools. Seonix publishes its own server: every read your dashboard can do — the Site Health report, the issue list with evidence, Search Console queries and pages, Google Analytics visits and sources, keyword clusters, articles, scores — becomes a tool the agent calls with your permission.

The agent does the thinking with its own model; Seonix supplies the facts and the scoring. No AI credits are spent on reads, and the agent never sees another account's data.

MCP URL https://api.seonix.ai/mcp · Read the agent manifest

Site Health checks every page for SEO, technical, speed and AI-search issues and marks which ones Seonix fixes for you.

What your agent can do

Six things to ask it today

Read Site Health and get a fix plan

The score by pillar, every open problem with the offending tag or redirect chain, and an ordered plan a coding agent can apply file by file.

Pull Search Console insights

Top queries and pages, the queries one page ranks for, striking-distance keywords at positions 4–20, indexing coverage and a 28-day trend.

Browse keywords and clusters

Researched keywords with volume and Search Console numbers, topic clusters with intent and page type, and the competitor insights behind each.

Score a draft before publishing

Any title, body and meta the agent wrote goes through the same SEO, readability and AI-answer checks the dashboard uses. Iterate until it passes.

Read articles and the calendar

What was published and when, each article's score and keyphrase, where it is live, and the slots planned for the month.

Follow Seonix playbooks

Eleven step-by-step procedures the agent can run: a full SEO review, promote the site, work through a page, fix issues in code, map keywords, write for a cluster, report monthly.

Try these

Three prompts to start with

Copy one into your agent once Seonix is connected. Each one maps to a playbook the server ships.

>Use Seonix: get the fix plan for my site, apply every page-level item in this repo, then tell me what to rescan.

Prompts are English on purpose: agents follow English instructions most reliably. Ask in your language — the agent answers in it.

Tool reference

46 tools, grouped the way the dashboard is

41 tools read and are marked read-only for the client, so agents do not ask before reading. The 5 write tools sit behind the write scope and the agent confirms each one with you. Every tool returns compact JSON with a hint on what to call next; 11 playbooks ship as MCP prompts.

01Account and project6

• get_account Projects, plans, quotas, whether the MCP is available
• list_projects Id, slug, name and site URL of each project
• get_project Autopilot mode, channels, Search Console status, links
• get_site_dna Brand, audience, tone and writing instructions as a prompt block
• get_plan_status Plan, permissions and the upgrade link
• get_seo_progress SEO Progress level and what the next one needs 02Site Health7
• get_site_health Score, pillars, counts, top open problems, last scan
• list_issues Issue threads by status, code or URL, paginated
• get_issue One issue with the captured evidence and the remediation profile
• get_fix_plan Ordered, implementable plan with URLs and target markup
• get_remediation_profile Why a code matters, bad vs good example, steps
• list_scans Recent scans with status and counts
• get_scan_status Progress of one scan 03Search Console8
• gsc_status Connection, property, last published day
• gsc_top_queries Top queries in the window
• gsc_top_pages Top landing pages in the window
• gsc_page_queries Queries one page ranks for
• gsc_opportunities Striking-distance keywords by estimated gain
• gsc_timeseries Daily clicks, impressions and position, 7–90 days
• gsc_coverage Indexed and not-indexed pages with verdicts
• gsc_impact Seonix's own impact estimate 04Google Analytics 45
• ga4_status Connection and GA4 property
• ga4_traffic Sessions, users, key events, visits per channel
• ga4_sources Every network that sent visitors, grouped
• ga4_pages Pages by visits with the source mix
• ga4_ai_traffic Visits from ChatGPT, Perplexity, Gemini and others 05Keywords and calendar4
• list_keyword_candidates Researched keywords with volume and GSC numbers
• list_clusters Topic clusters with intent, priority and winnability
• get_cluster Members, covering pages, SERP snapshot and insights
• get_calendar Slots for a month: kind, source, status 06Articles and results8
• list_articles Articles with scores and publish dates
• get_article One article with meta, takeaways, publications, HTML on request
• analyze_seo Score any draft: SEO, readability, AI-answer readiness
• score_article Re-score an existing article
• list_channels Connected publishing channels
• get_ai_visibility Latest ChatGPT-mention snapshot
• get_results Per-article clicks and positions vs the previous period
• get_blog_integration_guide How to connect a custom-built blog 07Playbooks2
• list_playbooks The eleven procedures with a one-line summary
• get_playbook One procedure in full, also served as an MCP prompt 08Fix and rescan (write scope, confirmed per call)6
• start_scan Start a fresh Site Health scan, then poll get_scan_status
• apply_fix Seonix applies auto-fixable items on WordPress through its plugin
• get_fix_run Reads a fix run: status and per-item outcome (a read, no scope needed)
• rollback_fix Undo a fix run on the live site
• acknowledge_issue Mark an issue as intentional so it stops counting
• unacknowledge_issue Count the issue again

Plans

Included in every paid plan

There is no separate MCP subscription: the server is another way into the project you already pay for, with the same limits as the dashboard.

On the Free plan nothing works: every call answers plan_required with the upgrade link. Write tools need the write scope you approve at sign-in, and the agent asks you before each change.

	Free	Starter and above
Connect an agent
Site Health, issues and fix plan
Search Console data
Keywords, clusters, articles, calendar
Draft scoring with analyze_seo
Playbooks and guides
Tool calls per agent	—	600 per hour

Safety

Built for agents that have shell access

Writes only after your yes

Reading is free of side effects. The five write tools (start a scan, apply or roll back Seonix fixes, mark an issue intentional) need the write scope you grant at sign-in, and the agent is told to ask you before each one.

Your data only

An OAuth grant or a key belongs to one account; every call re-checks that the project is yours and on a paid plan.

Untrusted content is labelled

Text crawled from your pages or taken from search results is marked as data, and the server tells the agent never to treat it as instructions.

Access you control

Every connected agent is listed under Account → Agents with its last use and client. Revoke it in one click. Access stops when you change your password.

Frequently asked questions

Which agents and clients work with it?

Any client that supports the Model Context Protocol over HTTP: Claude Code, Cursor, Codex, Windsurf, VS Code and others. Add the server by its URL and sign in with OAuth, or send an API key as a bearer header. Seonix is coming to the Claude and ChatGPT directories.

Does it spend my AI credits?

No. The tools return data and deterministic scoring; the thinking happens in the agent you run. Seonix spends AI only when you use its own generation features, which the MCP does not expose yet.

Is it included in my plan?

Yes, in every paid plan, with no extra charge. On the Free plan nothing works: the agent gets plan_required and the upgrade link.

How does the agent get access?

Through OAuth. You add the server by its URL or install the plugin; the first call opens app.seonix.ai, you see the client and the scopes it asks for and approve once. The agent receives a short-lived token and refreshes it itself. Revoke it any time under Account → Agents; changing your password revokes every grant.

Do I need an API key?

No. OAuth sign-in is the default and nothing is copied. Keys exist for clients that cannot open a browser: cron jobs, CI, scripts. Create one under Account → Agents, or let the agent obtain its own through the connect link described in the manifest; keys show up in the same list and are revoked the same way.

I have several projects. Which one does the agent use?

Pass the project id or slug, or connect with the per-project address /mcp/p/<project id> so every call is scoped to one site. With a single project nothing extra is needed.

Can the agent apply the fixes?

Two ways. In your own code: get_fix_plan says where and what to change, the coding agent edits the repository, deploys and calls start_scan to verify. On WordPress with the Seonix plugin: the agent calls apply_fix for the items Seonix can apply itself, you confirm in the chat, Seonix applies them and rollback_fix undoes a run.

What happens if the agent loops?

Each connected agent has a budget of 600 tool calls per hour. Past that the server answers with a clear error and a retry time instead of silently failing.

Can the agent connect my custom-built blog too?

Yes. The server serves the same integration guide as seonix.ai/SKILL.md, so a coding agent can build the publishing endpoints, test them and connect the channel from one session.

Connect your site, then connect your agent

Start with the free analysis. Once your project is on a paid plan, paste one line into your agent and it sees everything Seonix knows about the site.

Analyze my site free