Image Audit Tools With MCP Support

Image audit tools with MCP support can help an AI assistant find broken assets, inspect alternative text, and trace a problem back to the page that needs editing. The useful distinction is the evidence each connection returns. A list of image URLs, a page-level alt coverage count, and a browser-rendered image report answer different questions.

Screpy, Screaming Frog, Ahrefs, DataForSEO, and community tools offer different routes into that work. Choose by source-page traceability, crawl coverage, and retrievable fields before comparing issue totals. Through SEO MCP, the assistant can investigate connected data; the connection itself does not establish that every image was captured or that a proposed fix is correct.

This comparison uses vendor documentation checked in October 2026. It is published by Screpy and includes our product alongside alternatives. It is not a hands-on benchmark of every provider.

What an image audit needs to return

An image audit has two objects: the asset being requested and its use on a page. The asset might return a 404. Its use might have no alt attribute, an unsuitable description, or dimensions that need investigation. Repairing the file and repairing the HTML are different jobs.

Suppose /images/pricing-chart.png appears on a pricing page and an old blog post. Knowing that the asset is broken tells you what failed. Knowing both source URLs tells you where visitors encounter it and which templates or content entries need attention.

Before choosing a connection, ask whether you can retrieve:

• Asset identity: the image URL, including meaningful CDN transformation parameters.
• Page context: the source URL and, where available, the particular image occurrence.
• Original markup evidence: alternative text, whether the attribute exists, and declared dimensions when captured.
• Fetch evidence: response status and any available size or timing measurement, with units and measurement method.
• Capture context: crawl or run identifier, date, scope, and rendering settings.
• Completeness: pagination, page limits, and fields the tool did not collect.

These are selection criteria, not a promise that every product exposes all six categories. A useful report can state that transfer size is unavailable. It becomes misleading when an assistant replaces the missing measurement with an estimate and presents it as observed data.

Also check what a count means. Ten unique assets can appear in fifty page locations. A deduplicated asset inventory and an occurrence-level review can therefore have different totals while both are correct. Keep the two levels separate when assigning work or comparing tools.

Image audit tools with MCP support: six options

The options below range from saved website crawls to narrowly focused page checks. The table separates documented access from questions you should verify in your own connection.

Screpy: investigate images in a saved crawl

Screpy's MCP tool reference documents list_images for image records in one crawl and get_image_sources for pages referencing a selected image. That second step is useful when an asset appears across several pages and the team needs a repair location.

Select the project and crawl explicitly. Use structured filters to narrow the inventory, then follow source-page pagination before describing the affected scope. The result is evidence from that crawl. It does not prove that a recent edit is live, that every responsive variant was fetched, or that an image has been compressed. Those conclusions need additional inspection or a new crawl.

Screaming Frog: work with a configured desktop crawl

Screaming Frog's native MCP server provides access to SEO Spider crawl data and exports. It is useful when a team already works with local crawl files and wants an assistant to investigate them.

The Images tab documentation describes asset data and an Image Details view connecting an image to pages and alt attributes. It also distinguishes missing alt text from a missing attribute. For MCP work, ask for an export that preserves those relationships rather than assuming the top-level image table contains every occurrence.

Availability depends on the crawl's configuration. An export cannot recover evidence the crawl never collected. Confirm the required license and storage mode, then inspect the relevant report before treating the assistant's summary as complete.

Ahrefs: verify the image details inside Site Audit access

Ahrefs documents a Site Audit MCP tool group covering projects, issues, page explorer, and page content. Separately, its image-audit help page describes checks for image size, load time, content types, broken images, and alt attributes.

Those documents support a Site Audit investigation route. They do not establish that every image-report column is available in every MCP response. Inspect the connected tools and a representative result. If the response supplies an issue count but lacks the source URL or asset detail needed for a repair, retain the limitation and use the appropriate report to finish the investigation.

DataForSEO: join resource records to page context

The current DataForSEO MCP server uses documentation-discovery tools and an authenticated api_request. Older examples built around named tools should be checked against the active version.

For an existing OnPage task, the Resources endpoint supplies resource data, while Pages By Resource identifies pages containing a selected resource. This suits a workflow that joins API records into an audit queue.

Inspect the response configuration as well as the endpoint schema. A configured field selection can remove details an assistant needs. Preserve task identity, resource URL, source-page results, and pagination instead of assuming a single request supplies the complete relationship.

Apify: choose a specific image-audit Actor

Apify offers an MCP setup for smart-digital's image audit Actor. This is a community-maintained tool, not a description of every Actor on the platform.

Its documented output contains page records with image URLs, alt text, sizes, dimensions, and issue labels. Its browser-based capture can be useful for rendered pages, but the documented scrolling, size-measurement, and alt-classification limits matter. Review those limits before turning the issue count into editing instructions.

mcp-seo-audit: a lightweight HTML check

The open-source mk-techi mcp-seo-audit server documents audit_page with image alt coverage and crawl_site for aggregated page findings. Its core tools use public-page fetches without an API account.

That can fit a quick check of server-delivered markup. It is narrower evidence than a demonstrated inventory of fetched image assets, their transfer sizes, and their rendered dimensions. Request the actual output before selecting it for those tasks; an alt coverage result alone does not establish that broader capability.

Why source pages and alt text belong together

Alternative text belongs to an image's use in a page. Grouping a repeated file is useful for asset repairs, but it should not collapse every use into one universal description.

Consider an illustrative example: the same illustration appears beside an explanatory paragraph on one page and acts as the only content of a link on another. The first use needs to be assessed alongside the nearby explanation. The second needs an accessible description of the link's purpose. An asset URL alone does not tell you which situation applies.

The W3C alternative-text decision tree makes that context explicit. Decorative or redundant images can appropriately have empty alt text. Informative images need an alternative conveying their meaning; functional images need to communicate the relevant action or destination.

Preserve these states separately when the evidence allows:

A missing-value convention in a response may not preserve the difference between an absent attribute and an empty string. Ask for raw markup or explicit attribute presence where needed. Never infer that distinction from an aggregate count alone.

DataForSEO documents a related context limit: when the page URL is not supplied, OnPage Resources metadata is based on the page where the crawler first saw the resource. That is not a safe substitute for reviewing every occurrence. Obtain the relevant source pages and, when needed, page-specific evidence before recommending changes.

The practical requirement is a traceable join: asset, source page, captured alternative, and uncertainty. Without it, a tidy report can send an editor to the wrong page or replace a deliberately empty alternative with unnecessary text.

Check responsive and rendered-image coverage

Finding an <img> element does not establish which asset a browser loaded. A srcset attribute can describe several candidates, while the selected resource depends on the page's conditions. A flag showing that srcset exists is useful inventory evidence, but it is not a record of the selected image at a particular viewport.

Ask whether the connection inspected server HTML or a rendered page, which viewport it used, and whether it scrolled or triggered lazy loading. If you need browser-selected resource evidence, request it explicitly and keep it unavailable when the tool does not provide it.

Screaming Frog's crawl and MCP configuration guide documents the licensed MCP feature and database storage requirements. Its rendering and image-extraction settings also affect available evidence. Record those settings alongside the export; do not compare an HTML-only crawl with a rendered crawl as though their discovery scope were identical.

The Apify Actor discussed above documents that it does not scroll the page and derives remote image size from response headers. A lazy-loaded image can therefore be missed, and a size can remain unknown. An unknown byte value is not a zero-byte file or evidence that optimization is unnecessary.

Finally, separate a tool's warning threshold from an SEO verdict. Google's image guidance recommends discoverable HTML image elements, a src fallback for responsive images, and a balance between speed and visual quality. It supports formats including JPEG and PNG. A format label or fixed size threshold alone does not prove an image violates a Google requirement.

Use those warnings to shortlist assets. Then inspect their purpose, displayed size, actual delivery, and performance context. A small decorative icon and a large explanatory diagram need different decisions even when a tool groups them under the same issue label.

A small test before choosing an image audit tool

Use a small page set you control to check the connection before relying on it for a site-wide image audit. The fixture below is a proposed acceptance test, not a measured result for any vendor.

Inspect the test pages independently, then compare what the connection actually returns. Record the client, tool version, crawl settings, and date. Repeat the same scope when comparing two products. Mark an unavailable field as unavailable; do not let the assistant reconstruct missing evidence from a filename.

Issue labels need a check too. The Apify Actor's documented ALT_MISSING label combines missing and empty alternatives, and its background-image checks can produce alt warnings. Neither label alone proves that adding an alt description is the right repair. Evaluate the page's markup and purpose.

A connection passes your use case when its result identifies the affected asset and editable page, preserves the evidence you require, and reports its limits. It can still be useful if it needs a browser check for the final step. A connection that hides that dependency is harder to trust than one that states it clearly.

Retain the fixture as a regression check when changing tools, rendering settings, or response-field configuration. A connection that worked previously can become less useful if a necessary field disappears from its output.

Turn the results into a repair queue

Once the connection produces usable evidence, keep the investigation bounded. For Screpy, select one completed crawl, list the relevant image records, and retrieve their source pages. Investigate a manageable set before expanding the query.

This illustrative prompt describes the desired report; it is not a vendor response schema:

Use the selected project and completed crawl only.
Find image findings that need investigation and retrieve their source pages.
For each finding, keep the asset URL, source URL, captured status,
available alt evidence, crawl identity, and any missing fields.
Separate asset failures from page-specific markup questions.
Follow pagination or clearly label the result as a sample.
Return a proposed repair queue. Do not change content or settings.

Route a broken shared asset to the person responsible for its delivery or references. Route a page-specific alternative-description issue to the content or template owner. Where the capture was incomplete, make further inspection the next action instead of prescribing a speculative fix.

After a repair, inspect the live page and run a comparable follow-up check. Screpy's Website Audit workflow helps connect crawl findings to that verification step. A past crawl remains a past observation until new evidence confirms the change.

Choose saved-crawl access when traceability to existing project evidence is the priority. Choose a configured local export, an API integration, or a rendered-page tool when its documented output fits your requirements better. A lightweight HTML check may be sufficient for a narrower markup question. The deciding factor is whether the image audit tool with MCP support can show the evidence your team needs to act correctly.