Agent Utilities

Free static SEO release check

Review SEO changes
before your next release.

Check a page’s HTML, save a baseline, and compare it after your next real edit. Review new noindex directives, missing titles, canonical conflicts and observed broken links with the source evidence beside them.

No account, key or payment. The browser checks run locally. Only supplied static HTML is inspected; live HTTP status, search indexing, rankings and rendered JavaScript are not verified.

1. Check one page you are working on

Use a public page you own or are allowed to inspect, or sanitized HTML. Open the page’s View page source, copy the HTML, and paste it below. You can also load a saved HTML file. Entering a URL does not fetch it.

Your HTML stays in this tab. It is not uploaded or stored by this page.

Have a crawler snapshot JSON, or want a fictional example?

Load a capture from the local crawler or paste its JSON. A snapshot can contain up to 20 same-site pages. The fictional example is only a demonstration.

No audit run yet.

2. Review the evidence and save your baseline

The compact report and CSV omit raw HTML. The baseline contains your original HTML: keep it private and save a separate after-edit file. Pasted HTML has no verified HTTP status or source capture time.

Inspect finding details as JSON
Select an issue group after the audit.

After your next actual edit, check the same page again and save the new baseline under a different filename. Then compare the two files below.

3. Compare before and after your edit

Choose two saved baseline or crawler snapshot JSON files from the same site and page scope. This compares their stored evidence without fetching any pages. Each file can be up to 2.1 MB. Plain HTML files belong in step 1.

No snapshot loaded

No snapshot loaded

No comparison run yet. The fictional example has invented dates, HTML and HTTP observations.

Complete comparison, capture coverage and page availability

Newly observed does not always mean newly introduced. A finding missing later is not a confirmed fix: a failed read or different crawl coverage can hide it. Review capture dates and pages with unknown comparisons before acting.

Bring one real release check

We can help set up a free baseline for 5–10 pages you are permitted to inspect, then help compare them after your next real edit. Start with one question: Which check do you repeat after edits? We can establish a permitted public or sanitized scope and your usual review frequency in the reply. No credentials or private client data.

Tell us your recurring check

Already tried it on real work? Share the task and outcome. Feedback is optional. Setup help is manual; this is not scheduled monitoring.

Capture a small permitted site with the local crawler

The local crawler pins validated public DNS addresses, stays on one HTTPS origin, respects robots.txt and waits at least one second between requests. Optional XML sitemap discovery includes same-origin unlinked pages, with up to five sitemap requests and 200 candidates. It follows no off-site canonical or sitemap URLs. Limit: 20 pages, 100 KB per response, 2 MB total and 60 seconds. Missing, blocked, failed and unvisited pages remain unknown. Only run on sites you own or have permission to inspect.

Download Node.js 22+ crawler · SHA-256 checksum · Dependency licenses

shasum -a 256 -c agent-utilities-seo.mjs.sha256
node agent-utilities-seo.mjs crawl https://YOUR-SITE/ --out crawl.json --max-pages 10 --sitemaps
node agent-utilities-seo.mjs refresh crawl.json --out refreshed.json
node agent-utilities-seo.mjs changes crawl.json refreshed.json --out seo-changes.json
node agent-utilities-seo.mjs report crawl.json
node agent-utilities-seo.mjs details crawl.json --group noindex --offset 0 --limit 20

The offline changes command compares stable SEO findings from two bounded captures, retaining dates, coverage and unknown page observations. Findings not observed later are not resolved issues; changed crawl context can hide them. It saves a private file and refuses existing output. The crawl snapshot contains raw HTML. Keep that file out of model prompts; give your agent the compact report and bounded details. Summary and detail output are measured in bytes here, not tokens. Small inputs may produce larger reports.

Try an SEO comparison before crawling

This fictional offline example contains invented HTML, HTTP statuses, coverage and capture times. No site was fetched and no customer task is represented. Download the files into the same folder as the crawler:

Before snapshot · After snapshot · Complete expected result · Checksums · Machine-readable example manifest

shasum -a 256 -c seo-change-example.sha256
node agent-utilities-seo.mjs changes seo-change-example-before.json seo-change-example-after.json --out my-seo-changes.json
cmp my-seo-changes.json seo-change-example-result.json

The home page gains a title and a newly observed noindex directive. The contact page changes from a supplied 404 observation to a failed 503 observation; its metadata comparison stays unknown. The missing-title finding is no longer observed, which does not by itself prove a resolved issue. Both captures cover two of three invented discovered URLs. The result retains the capture times and coverage, with no raw HTML. The command makes no network or model request, saves a private file and refuses an existing output path. A successful comparison prints nothing from cmp.

Reuse pages without sending the whole site

Prepare a local compact library from a crawl snapshot, or collect up to 20 bounded Wikipedia articles from one language edition. Search the stored text for your current task and give the selected output to your agent. This is lexical matching with explicit coverage and omissions; it is not a complete website archive or semantic search.

Download local page library · Checksum · Dependency licenses

node agent-utilities-pages.mjs prepare crawl.json --out library.json
node agent-utilities-pages.mjs search library.json --query "pricing changes" --max-chars 6000 --out selection.json
node agent-utilities-pages.mjs compare library-before.json library-after.json --out changes.json
node agent-utilities-pages.mjs wikipedia https://en.wikipedia.org/wiki/Earth --out earth.json
node agent-utilities-pages.mjs search earth.json --query "atmosphere" --out selection.json

Wikipedia cache reuse avoids another source request for 24 hours. Use --refresh to fetch the current article again. Revision and license travel with selected text. Extraction is bounded at 12,000 characters, not a full article mirror. Generic libraries have at most 20 supplied pages. Compare two saved libraries to see changes in text, titles and source versions before rereading them. Capture dates and omissions stay explicit; a missing observation does not prove deletion. Comparison does not fetch. The crawler refresh command rechecks the exact previous URL set and emits compact changes, preserving the original snapshot. Refresh is manual; no background scheduler runs. Snapshots stay local; nothing is submitted to archive.ph or publicly hosted.

Inspect a dated output before setup

Read the complete Moon section example · Checksum. This fixed result selects Names and etymology from Wikipedia revision 1378452949, captured October 6, 2026 at 21:48:36 UTC. Attribution, license, revision links and omissions are retained. It is a public-source example, not a live read or user task. No private library or raw HTML is shared. This is a selected result, not a collection file for serve or an MCP config.

The complete JSON file is 6,395 bytes and 1,556 o200k_base tokens with tiktoken 0.14.0. Its selected section is complete within the captured text, but the article capture is partial. These are file counts, not billed input, Claude/Qwen counts or dollar savings. Choose a relevant source for your own task.

Collect and select Wikipedia articles

Download the Earth/Moon starter list. This is an example collection, not a popularity ranking. Replace the URLs with your reading task.

node agent-utilities-pages.mjs collect wikipedia-collection-example.json --out collection.json
node agent-utilities-pages.mjs search collection.json --query "orbital period" --max-chars 4000 --out search-selection.json
node agent-utilities-pages.mjs select collection.json --url https://en.wikipedia.org/wiki/Moon --section "Names and etymology" --max-chars 4000 --out moon-selection.json

Each page retains its retrieval date, revision, license and omission flags. Fresh per-page cache reuse lasts 24 hours. Use collect ... --refresh to request current reads. Failed or skipped refreshes can retain prior text with an explicit stale status. Source reads share 8 MB and 60 seconds. Exact section selection uses captured headings; an omitted or ambiguous heading is reported rather than guessed. Search and selection do not fetch. The collection stays in a private local file.

Connect your agent to the saved collection

The same module is a local MCP server. Generate a connection file from your saved capture; the command validates it and resolves the current Node executable, module and capture paths automatically.

node agent-utilities-pages.mjs config collection.json --out saved-pages-mcp.json

Copy or merge the generated server entry into your stdio-capable client. This command does not edit client settings, install a connection, fetch sources or upload the capture. It refuses an existing output file. The file contains local paths, so keep it private. The manual connection template is also available.

node agent-utilities-pages.mjs serve collection.json

Your client can list captured page metadata, search selected text, select an exact captured section and inspect stored outlines. It receives only the requested result, with revisions, dates, stale status and omission flags. The server loads one fixed capture; refresh it separately and restart the connection to load new data. This local connection does not fetch sources, read client-selected files, call models or make payments. The four saved-page tools are distinct from the hosted service tools.

Use --out selection.json with search or select to save the complete payload with private file permissions and preserve any existing file. Without --out, these commands emit the payload on stdout; shell redirection uses shell permissions and can overwrite files.

Measure actual tokens for a named encoding

Download token measurement script · Checksum. Install the open source tokenizer in your own Python environment:

python3 -m venv .tokenizer-env
.tokenizer-env/bin/python -m pip install tiktoken
.tokenizer-env/bin/python agent-utilities-tokens.py selection.json --encoding o200k_base --cache token-cache.json

The count includes the complete selected JSON file. Pretokenizing unchanged text does not reduce its tokens. Counts and optional IDs are specific to the selected encoding; they are not universal counts for Claude, Qwen or every model. Cached tokenization saves local processing; smaller relevant text may reduce model input. No token-to-dollar claim is made.

Use the same audit from your agent

Free HTTP: POST /v1/seo/audit, JSON snapshot up to 128 KiB. Hosted MCP: agent_utilities_seo_audit at /mcp/remote?category=web. The free audit does not call paid utilities, purchase credits or crawl URLs remotely. Supplying full snapshots in an MCP request adds input tokens; the local crawler/report path keeps HTML outside the model.

Schema, example and detail pagination · MCP setup · Other free utilities