Claude SEO v2.6.2 is a community release dated in the project changelog. GitHub records it as published at 10:10 UTC that day, and the tag resolves to commit 468ed5b. It follows v2.6.1, a security release published on November 14, 2026 at 14:25 UTC, and v2.6.0 of November 10, 2026. v2.6.2 is the current public release as this post is published.
In short: update. v2.6.1 fixed the problems behind ten private security reports, and v2.6.2 contains those fixes plus two new extensions (/seo shopify and /seo posthog), a way to trust a local proxy, and 36 community pull requests. The plugin now ships 29 sub-skills, 20 agents and 11 extensions. A few defaults and exit codes changed. They are listed below.
What did v2.6.1 fix?
v2.6.1 is a security release with no new commands. Eight researchers sent ten reports through GitHub's private advisories, and all of them are fixed. The ones that matter most:
- The headless browser can no longer be redirected to local addresses. The earlier guard checked only the first URL of a request. An audited site could redirect Chromium to a loopback, private network or cloud metadata address and have the response returned in the audit. Every browser connection now goes through a local guard that checks each destination, including every redirect. Unlighthouse runs behind the same guard. Reported by namaserajesh and by Nordalux, who also wrote the initial fix.
- Chromium runs with its sandbox wherever the system can start it. Playwright starts Chromium without a sandbox unless asked. It is now asked for first. Reported by stephaneRR and shayanabbas.
- Report values are escaped, and the PDF can load only its own images. Search Console queries and other API text went into the report HTML as they came, and the PDF renderer would fetch any file or URL the markup named. Reported by samnguyen80 and namaserajesh.
- The Excel export stores text as text. A search query that starts with
=,+,-or@can no longer run as a formula. Reported by 0xGenesi. - Google sign-in asks for read-only access. The Indexing scope is granted only on request, and the unused Search Console write scope is gone. Reported by namaserajesh.
- Smaller fixes:
youtube_search.pyremoves the API key from its errors (georgekhananaev),verify_release.pyrefuses manifest entries outside the checkout (tangzheng202202), andparse_html.py --urlreads under a size cap and a time budget (samnguyen80).
The advisory IDs are in the v2.6.1 release notes. Thank you to everyone who reported these privately and waited for the fix.
What behaves differently after v2.6.1?
- The Chromium sandbox is best effort. On systems that cannot start it (Ubuntu 23.10 and later, GitHub's Ubuntu runners, most Docker images, root) pages still render, a warning says why, and the JSON result carries
sandbox: false. SetCLAUDE_SEO_CHROMIUM_REQUIRE_SANDBOX=1to fail instead. - A new Google sign-in is read-only. It asks for read-only Search Console and Analytics access. To use
indexing_notify.py, sign in withgoogle_auth.py --auth --indexing. Existing sign-ins keep working with the access they were granted, so sign in again if you want the narrower access. - Browser commands refuse a proxy on a local or private address, and SOCKS or
https://proxy URLs. v2.6.2 adds an opt-in for a local proxy you trust (see below). - A failed render is an error. A render, screenshot or visual analysis that ends on a blank page, a browser error page or a blocked address is reported as an error instead of being returned as the page.
What is new in v2.6.2?
- Shopify extension (
/seo shopify, #302 by Nordalux). A sitemap-complete crawl of a store you administer, signed with the store's Crawler Access signature from a project-local.shopify-env, then handed to the audit. A signed crawl has no page cap. Without a usable signature the crawl runs at the unsigned limits: 500 pages, concurrency 2 and a 1 second delay. robots.txt is respected by default, signed or not. - PostHog extension (
/seo posthog, #336 by AlexMili). Organic traffic, top landing pages, device and country splits and referrers from the PostHog MCP server, as a GA4 alternative or complement. There is no script and no stored credential. The agent runs read-onlySELECTqueries only. The setup guide recommends a read-only PostHog login or API key, because the MCP tool itself can also write. - Local proxy opt-in (#398 by eduardopicazzo).
CLAUDE_SEO_TRUSTED_PROXIESnames a proxy on loopback or a private address that you trust, ashostorhost:port. Fetches, screenshots, rendering and Unlighthouse then work in Claude Code cloud sessions and behind tools such as cntlm or px. Every destination is still checked before the proxy is asked for it. Link-local and cloud metadata addresses stay refused even when listed. Without the variable nothing changes. - Per-project credentials (#341 by lciolecki).
CLAUDE_SEO_PROFILE_DIRpoints the Google, backlink and Matomo credential files at a per-project directory. The scripts warn when that directory sits inside a git work tree, where keys could be committed by mistake. /seo bing,/seo shopifyand/seo posthogload on plugin installs. The three skills now ship with the plugin, and their installers only do configuration there. A plugin-only install had none of them before.
What did the community fix?
v2.6.2 merged 36 pull requests from 8 contributors. 29 of them are correctness fixes by Afloat16 across reports, drift, schema, transport grading, DataForSEO, preload, sitemap, CrUX, Search Console, IndexNow, parasite risk and domain history. Some of what changed:
- CrUX trends compare the first and last four calendar weeks, so old measurements are not presented as recent.
- Drift compares Core Web Vitals only when both snapshots hold field data for the same scope. That removes false regressions when coverage changes.
- Batch URL inspection counts only the URLs it actually inspected, so indexation rates are correct.
- The CSP grade follows the directives browsers actually apply to scripts, and a duplicate HSTS directive is flagged as invalid.
- DataForSEO Merchant labels such as "sold out" and "unavailable" are never read as in stock.
- PDF reports work on macOS with Pango installed through Homebrew (#457 by Vivek-Chaudhari30).
content_quality.pycounts whole words in Greek, Cyrillic, Arabic, Hebrew, Devanagari and other scripts that put spaces between words (#455 by rampilea).- The UCP report keeps the profile URI of each supported version (#454 by kevin9327).
- On Windows, a certificate that chains to a valid root no longer fails verification. This one came from a bug report (issue #421), not a pull request.
Every fix is listed in the changelog. Thank you to everyone who sent one.
What changes if you script against the tool?
- IndexNow requires an exact host match. A URL on a subdomain of the declared host is no longer submitted.
- Exit codes.
preload_check.pyexits with code 1 when the page cannot be fetched or answers with an HTTP error. It scored the error page before.parasite_risk.pyexits with code 1 and reports unknown risk when no page could be audited. That case used to pass as low risk. seo_updates.py --limit 0means no limit. It returned an empty list before.- Drift baselines for URLs with path parameters (
;name=value) need capturing again. - Audit reports are named
SEO-Audit-Report-*. Reports built from Google API data keep theGoogle-SEO-Report-*name.
What do the counts look like now?
| Item | v2.6.0 | v2.6.2 |
|---|---|---|
| Sub-skills | 26 | 29 |
| Specialist agents | 19 | 20 |
| Extensions | 9 | 11 |
| User-invocable commands | 34 | 36 |
| Python scripts | 63 | 67 |
| Automated tests | 2,358 | 3,902 |
The skill, agent, extension and script counts come from the repository at each tag. The three new skills are seo-bing, seo-shopify and seo-posthog, and the new agent is seo-posthog. The two new commands are /seo shopify and /seo posthog. The test counts are the tests collected in CI at each tag.
How do I update to v2.6.2?
Reinstall through the plugin marketplace, or run the installer pinned to the tag, then check the runtime. The installation guide covers Windows and manual installs.
curl -fsSL https://raw.githubusercontent.com/AgriciDaniel/claude-seo/v2.6.2/install.sh | bash /seo doctor /seo page https://your-site.com
The Python dependencies are the same as in v2.6.0. If you come from an older version, run /seo setup once. The Shopify and PostHog extensions each have their own installer in the repository's extensions folder.
How was the release checked?
The v2.6.1 and v2.6.2 pull requests each passed every CI check (31 check runs), including the test suite on Python 3.11, 3.12 and 3.14, a browser guard test on Linux, macOS and Windows, a dependency audit and a secret scan.
On November 17, 2026 the repository had 30,035 stars and 3,780 forks.
Official source: Read the Claude SEO v2.6.2 release for the canonical summary and tag.
