The best Codex plugins for online marketing: SEO, ads and analytics

What a Codex plugin is in 2026

OpenAI is explicit that ChatGPT and Codex "use one universal plugin directory, so the same public plugins are discoverable from their supported surfaces." A plugin bundles skills, MCP servers, browser extensions and hooks. In Codex, plugins work in Codex inside the ChatGPT desktop app (the separate Codex desktop app has become a mode of the ChatGPT one) and in the Codex CLI, where /plugins opens a plugin browser. The IDE extension does not support plugins. Next to the public directory there is OpenAI's curated "Codex official" marketplace on GitHub, with 65 plugins in November 2026, and vendors can publish their own marketplaces, which you add with codex plugin marketplace add owner/repo.

The second route is the one ChatGPT does not have. Any MCP server can go in ~/.codex/config.toml, or in a project's own .codex/config.toml once you trust the project, either as a local command or as a remote URL with a bearer token or OAuth through codex mcp login. The ChatGPT desktop app, the CLI and the IDE extension all read that file, while ChatGPT on the web does not. Per server you can limit the tools with enabled_tools and set default_tools_approval_mode, where writes asks before every tool that is not marked read-only. My Codex Search Console MCP setup shows the whole configuration for GSC Wizard.

Codex is included in every ChatGPT plan from Free upwards and shares its usage with ChatGPT Work. Free and Go get it in the desktop app only, as it rolls out; the CLI, the IDE extension, the web and iOS start at Plus, which costs $20 a month. By default Codex works with network access turned off, and the Auto preset asks before it touches the network or edits outside the project folder. MCP tools have their own rule: "Destructive app/MCP tool calls always require approval when the tool advertises a destructive annotation."

Google has no plugin in the directory, but in Codex that matters less than in ChatGPT, because Google's own servers run locally. The Google Analytics MCP server and the Google Ads MCP server both work in Codex, and Google documents the Codex setup for the Ads one. Google publishes no Search Console MCP server, so Search Console data still comes through a third party.

How to judge a Codex marketing plugin (five questions)

The shortlist at a glance

Rank = position of 13,550 plugins in the shared ChatGPT and Codex directory, per mcptop.com on 13 November 2026. Local servers are not in the directory.

SEO

The same split as everywhere: tools that read your own Search Console data and tools that sell estimates about the whole web. What Codex adds is that the answer can become a script, a CSV or a pull request instead of a paragraph.

1. GSC Wizard

Good at: your own Search Console, GA4 and Bing Webmaster data, analysed server side against a warehouse that keeps years of history, well beyond Google's 16-month limit. In Codex that turns into work you can keep: ask it to find the pages that lost non-brand clicks since the last core update, then write the list into the repository and open each template that needs fixing. It is the highest-placed plugin built on first-party Search Console data in the shared directory, at #60 overall and first in Data & analytics.

Watch for: it is a paid product, and it is mine. It does not estimate competitor traffic or backlinks. The plugin and a server in config.toml are two separate connections: the plugin uses the connection on your ChatGPT account, so switching GSC Wizard accounts means disconnecting it in ChatGPT, not only in Codex. A handful of tools change things, such as IndexNow submissions, so set the server's approval mode to writes.

Access: plans from €17/month with a 7-day free trial. Install it from the directory (GSC Wizard plugin), or run codex mcp add gsc-wizard --url https://mcp.gscwizard.com/mcp and then codex mcp login gsc-wizard.

2. Ahrefs

Good at: backlinks, keywords, rank tracking and site audits from Ahrefs' hosted server at https://api.ahrefs.com/mcp/mcp. A ?tools= parameter on that URL limits which tool groups load, which keeps Codex's context small when you only need Site Explorer.

Watch for: the data is modelled and every call consumes API units; keys created for MCP default to 200,000 units a month. Ahrefs does not document Codex, and its terms say that calling the MCP endpoint "via custom scripts, bridges, or standalone HTTP/JSON-RPC clients is unsupported and not permitted." Let Codex use Ahrefs as a tool, and use the regular Ahrefs API for any report script it writes.

Access: paid Ahrefs plan from Lite. Ahrefs in the directory.

3. Semrush

Good at: keyword research and competitive analysis on Semrush's own database, through https://mcp.semrush.com/v2/mcp. Semrush also publishes an official set of agent skills built on that server, and its README names Codex among the supported tools.

Watch for: it needs a Semrush One or SEO Classic plan with API units, and every request consumes them. Your projects are read-only through MCP: you "can get data from your projects this way, but not create new projects, modify settings, or perform other actions." Semrush's setup pages cover other clients but not Codex.

Access: Semrush subscription with API units. Semrush in the directory.

4. SE Ranking

Good at: being the major SEO platform that documents Codex properly. One command, codex mcp add se-ranking --url https://api.seranking.com/mcp, adds the server and starts the sign-in, and the same configuration works in the CLI and the IDE extension. It covers keywords, backlinks, SERPs, audits and SE Ranking's AI visibility data.

Watch for: the data available depends on your subscription, and not every API endpoint is exposed through MCP. Sub-accounts cannot use the sign-in button and need an API key header instead.

Access: SE Ranking API plan; every successful call uses credits. SE Ranking in the directory.

5. DataForSEO

Good at: raw SEO data for people who build their own tools: keywords, SERPs including Google AI Overviews, competitors, backlinks, on-page audits and local results. DataForSEO publishes an official Codex plugin (codex plugin marketplace add dataforseo/codex-plugins) that connects to its hosted MCP server, and the sign-in opens automatically during installation.

Watch for: it is pay-as-you-go per request, and the cost "depends on the selected API, endpoint, parameters, and workload". That suits Codex well, until Codex writes a loop over 10,000 keywords. Put a hard cap in any script that calls it.

Access: DataForSEO account with credit. DataForSEO's Codex plugin.

Google Analytics, Google Ads and paid media

This is where Codex pulls ahead of ChatGPT. Google's own servers run on your machine, so GA4 and Google Ads data can reach Codex without a third party in between, and the multi-platform vendors cover everything else.

6. Google Analytics MCP

Good at: GA4 straight from Google. The server runs standard, funnel and realtime reports, and a conversions report with ad costs, conversions, ROAS and attribution models, using the Analytics Admin and Data APIs. It reads with the read-only Analytics scope by default.

Watch for: setup is the price. You need a Google Cloud project with both APIs enabled, Application Default Credentials and Python with pipx, and Google's README documents other clients but not Codex, so you add it as a local command yourself. The README now also lists three write tools (annotations, custom dimensions and key events) behind an --allow-write flag, while Google's developer guide still calls the server read-only; leave the flag off unless you need it.

Access: free; a Google account with access to the GA4 property. Google Analytics MCP on GitHub.

7. Google Ads MCP

Good at: Google Ads data through GAQL queries, with a documented Codex setup: run the server locally, then codex mcp add google_ads --url http://localhost:8080/mcp and codex mcp login google_ads. It can also create assets, upload videos and manage labels, each through a propose-and-confirm step that rejects a change whose checksum does not match.

Watch for: it has no tools for campaigns, budgets or bids, so optimisation still happens in Google Ads. Test access only reaches test accounts; production accounts need Explorer, Basic or Standard API access, and since developer tokens were retired on 9 September 2026 that access level comes from your Google Cloud project.

Access: free; Google Ads API access approved for your Cloud project. Google's integration guide.

8. Windsor.ai

Good at: cross-channel reporting from one connection to more than 350 sources, including Google Ads, Meta, Microsoft Ads, LinkedIn and TikTok. It is the highest-placed marketing plugin in the shared directory at #14, and it can act as well as read: pausing campaigns, changing budgets, creating ads. New Meta ads are created paused, and most pause, enable and update actions can be undone; creating something new cannot be undone in one step.

Watch for: Windsor.ai does not document Codex, so use the plugin or its generic remote server at https://mcp.windsor.ai/. Its pricing page says write actions are available on every plan, which means they are available to Codex too; set the server to writes approval.

Access: free plan with one data source and 30 days of history; paid plans from $19/month. Windsor.ai in the directory.

9. Adspirer

Good at: managing campaigns across Google, Meta, Amazon, LinkedIn and TikTok in plain language, with proper Codex documentation: its own plugin marketplace (codex plugin marketplace add amekala/ads-mcp) or the server at https://mcp.adspirer.com/mcp.

Watch for: it writes to live ad accounts, which is exactly what you do not want combined with a Codex session running without approvals. Adspirer's own guardrails help: every campaign is created paused, "even in autonomous mode", and budgets do not change without confirmation.

Access: free for 15 tool calls a month; Plus $49 for 150, Pro $99 for 600 and Max $199 for 3,000. Adspirer's Codex guide.

Product analytics

Both picks here document Codex and ship their own plugin, and both suit Codex for a second reason: the same session that analyses an event can add the tracking for it in your code.

10. PostHog

Good at: the funnel after the click, plus the instrumentation behind it. PostHog is in OpenAI's curated Codex marketplace, documents a one-line setup (codex mcp add posthog --url https://mcp.posthog.com/mcp), and its own plugin approves known read calls automatically while every write still gets Codex's approval prompt.

Watch for: it can create and delete feature flags, insights and experiments that belong to your product team. Add ?readonly=true to the URL and every tool that creates, updates or deletes data disappears from the list.

Access: connecting and calling the MCP tools is free; PostHog has a free plan. PostHog's Codex guide.

11. Amplitude

Good at: behavioural analytics and tracking plans. Amplitude documents Codex and publishes its own Codex plugin, described as "an expert analyst and instrumentation partner inside Codex", which can read charts and create or edit dashboards, experiments, flags, events and cohorts.

Watch for: every delete is a two-step action that asks you to confirm, and write calls need the "Use MCP (write)" permission in Amplitude, so an admin can keep a role on read access. The answers are only as good as your event names.

Access: any Amplitude account, over OAuth. Amplitude's Codex guide.

Technical SEO in the codebase

None of the chat assistants can do this part well, because the fix lives in code. Codex can read the site's templates, test the rendered page and change both.

12. Chrome DevTools MCP

Good at: measuring a page the way Google's tools do. Built by the Chrome DevTools team and documented for Codex (codex mcp add chrome-devtools -- npx chrome-devtools-mcp@latest), it runs Lighthouse audits for accessibility, SEO and best practices, records performance traces for LCP, INP and CLS with real-user CrUX data alongside, and can emulate a mobile viewport, a slow network or a crawler's user agent.

Watch for: the Lighthouse tool excludes performance, which comes from the trace tools instead. It drives a real browser, and attaching it to your own Chrome exposes your logged-in sessions to the agent, so leave --autoConnect off and use --isolated for a clean temporary profile. Usage statistics are collected by default; --no-usage-statistics turns that off.

Access: free and open source; needs Node.js and Chrome. Chrome DevTools MCP on GitHub.

Also worth a look

Three traps with Codex marketing plugins

The script that spends your credits

Ask Codex for keyword data on a whole category and it will often write a loop rather than make three tool calls. Against a metered API such as DataForSEO, Semrush or Ahrefs, that loop can spend a month's allowance before you read the output, and Ahrefs does not permit scripts against its MCP endpoint at all. Ask Codex to estimate the number of calls before it runs anything, and put a limit in the code.

Credentials in the repository

A project-level .codex/config.toml is convenient and easy to commit by accident, and an API key pasted into the chat ends up in the session history. Keep keys in environment variables (Codex reads a bearer token from one with bearer_token_env_var), prefer OAuth sign-in where the server supports it, and keep local credential files out of version control.

Full access mode with write tools

Codex has a mode with "no sandbox; no approvals", and its own documentation marks it "not recommended". Combined with an ads server that can create campaigns or a host plugin that can deploy, a misread instruction goes straight to production. Keep the default Auto preset for marketing work, and set default_tools_approval_mode to writes or approve for every server that can change an account.

Three setups that work

Frequently asked questions