Local SEO tools.
Inside your agent.

Connect SEO Crawl Audit to Codex or Claude Code. Ask your agent to plan a crawl, find SEO issues and save a report in your workspace. Start with the installation prompt for your client.

Native packages are separate from the npm tarball. Verify each download against its matching SHA-256 checksum.

SEO Crawl Audit's coral crab logo

Choose your agent.
Copy its prompt.

Two setup prompts for two workflows. Your agent checks the latest public release, installs it for your workspace and tests the connection. You stay on this page.

Codex

Install the native plugin through a local marketplace. Bind it to an existing workspace and a Node executable.

Read the installation prompt
Install or update SEO Crawl Audit in Codex using the latest public native release from https://github.com/MelnixDev/seo-crawl-audit/releases/latest. Respond in my language.

1. Inspect the existing Codex plugin/MCP registrations first. Ask for my existing audit workspace only if it is not known. Preserve unrelated configuration and uncommitted work.
2. Resolve the latest release tag, download its codex-claude-code ZIP and matching .sha256, and verify the checksum before extracting. Use a stable local marketplace outside the audit workspace; the directories must not contain each other.
3. Read the bundled INSTALL.md. Run configure.mjs with my audit workspace using the intended Node executable (Node >=20.19). Register the local marketplace and install through the supported Codex plugin commands. For an existing installation, follow its update/reinstall procedure. Avoid a duplicate legacy MCP registration; preserve its configuration before migrating it.
4. Report the installed plugin version and bound workspace. Tell me when a Codex restart/new session is required; do not claim the current session has reloaded automatically.
5. Verify discovery of all eight tools and the engine version in a fresh Codex session. Plan first, then run a maximum one-page scan against a controlled localhost fixture, respect robots.txt, inspect paginated issues and generate/open the HTML report. Keep artifacts in my audit workspace. Do not crawl a production site as an installation test.
6. Distinguish verified native-loader/tool results from GUI installation checks. Report concrete errors and the next recovery step. Never print secrets or overwrite unrelated registrations.

Claude Code

Use Claude Code in your terminal or the Code tab of the desktop app. The prompt selects the appropriate setup path and explains when a Desktop MCPB extension is needed.

Read the installation prompt
Install or update SEO Crawl Audit for Claude Code using the latest public release from https://github.com/MelnixDev/seo-crawl-audit/releases/latest. Respond in my language.

1. Inspect the current environment and existing SEO Crawl Audit registration. Claude Code can run in the terminal or the Code tab of the Claude desktop app. Establish which installation surface is available before changing settings; use the matching instructions in the public native installation guide.
2. Resolve the latest release tag and download the matching codex-claude-code ZIP plus .sha256. Verify the checksum before extracting. Read the bundled INSTALL.md. Keep the plugin outside the audit workspace and preserve unrelated settings.
3. For the Claude Code plugin path, use the documented local marketplace/plugin directory workflow and bind the existing project directory. Do not assume Codex configuration applies to Claude. If this environment instead requires a Desktop extension, use the public claude-desktop.mcpb and its matching checksum, explain that the bundle is unsigned, and follow the Desktop workspace picker flow. Do not present the ZIP and MCPB as interchangeable.
4. Follow the documented update/reinstall flow for an existing installation and avoid duplicate server registrations. Tell me when restart or a manual install/permission step is necessary; do not claim an unfinished step passed.
5. Confirm discovery of all eight MCP tools and the installed engine version. Plan first, then scan a controlled localhost fixture with maxPages: 1 and robots.txt respected, inspect paginated issues, generate and open the local HTML report. Do not crawl production during setup.
6. Keep artifacts in my chosen workspace. Report actual client/version, what passed and what remains unverified. Never print secrets or change unrelated client configuration.

Installation may need client permissions or a restart. Prompts guide the setup; they do not bypass client approval. Manual downloads and checksums · Full installation reference

Codex, checked
with a real run.

Public plugin 0.17.4 was installed through Codex’s native marketplace importer on macOS. A fresh Codex app-server loaded all eight tools and confirmed engine version 0.17.4.

The test planned and crawled one localhost page, read its issues and generated an HTML report. It used Codex CLI 0.169.0-alpha.11.1 and Node 24.8.0 on 14 November 2026.

  1. Install and restartVerify the download, configure the workspace and install the plugin. Start a fresh client session to discover its tools.
  2. Check the connectionConfirm the engine version and all eight tools. Test one controlled page before selecting a production site.
  3. Inspect the resultRead the affected URLs and open the local report. Review scope and coverage before acting on a finding.

Verified: native installation, new-process tool discovery and bounded tool workflow. Interactive Codex GUI installation/update/removal is still tracked in the Codex acceptance task. Claude client acceptance is tracked separately.

Eight tools.
One audit trail.

Plan and scan, inspect issues in pages, compare runs and open a local report. The plugin uses the same report interface as the CLI. Explore the report example.

Know what was verified.

Runtime and client boundaries

The published npm CLI and MCP workflow have been exercised against real downloads. Native-package schema and archive checks verify packaged paths and bytes. Native Codex CLI loader testing verifies skill icons, all eight tools and an actual bounded scan with an empty inherited PATH.

Interactive GUI installation and Claude client workflows remain separate verification steps. No public marketplace listing or platform approval is implied.

Artwork and local artifacts

Packages carry the same crab artwork in client metadata. Support differs by client: Claude Code’s directory icon field is ignored at runtime. Metadata validation does not prove every picker renders an icon.

The crawler requests your chosen site. Audit artifacts stay in the configured workspace, while the AI client handles tool results under its own policies. Read branding support notes.

What does a crawl tell me about indexing?

It reports observed URLs and rule findings within the selected scope. It cannot establish Google indexing, rankings or complete site coverage. Review heuristics in context and use Search Console for indexing evidence.

Does the package download a browser or install npm dependencies at runtime?

The native archives contain their JavaScript runtime dependencies. Optional browser rendering requires Playwright and Chromium installed separately in the consumer workspace. Default scans use returned HTML.

Can I audit a different workspace after installing?

A configured Codex copy stays bound to its explicit workspace. Use the documented configuration and reinstall flow to change it. Workspace and plugin directories must not overlap. The server rejects calls that name another workspace.

Start small.
Ask for evidence.

Plan an audit of my site. Show the scope first, then scan up to 100 pages while respecting robots.txt. Return the highest-priority findings with URL evidence and the HTML report path. State coverage limits.

For Codex, include the configured absolute workspace. A prompt does not override your client’s tool-confirmation policy.