An MCP server for accessibility.

Let Claude and other AI agents check your sites, work through fixes and report back, inside limits you set. 30 tools, included with AI Fix.

Start free See AI Fix

MCP, in one paragraph.

MCP is the standard plug that lets an AI tool use other software. Connect Scanpoof and your agent can read your scan results, check a page on the spot, ask Scanpoof's AI for fixes and see each one tested on your live page. There is no chat window in Scanpoof: the agent is whatever you connect, and what it can touch is set on our side.

Works with any MCP client that can send an Authorization header
Nothing goes live unless you allow it
Every action logged under the key's name, and for whom
No separate fee: it comes with AI Fix

What a connected agent can do.

30 tools in three groups. Looking and checking never change your site. Acting is limited by the key and, for anything that goes live, by a switch only you control.

The server also offers ready-made prompts, and every rule Scanpoof reports as a readable resource, so an agent can look up what a rule means before it fixes anything.

Look (17)

list_sites: Your sites, each with its score and how many fixes are waiting, live or need a person.
get_scan_summary: The latest scan: score, issues by severity, the most common problems.
get_most_urgent_issues: The issues that matter most, ranked by severity times the pages they appear on. A header problem repeated on every page counts once.
list_issues: Issues from the latest scan, worst first, with the element and the page it is on.
explain_rule: What a rule means in plain words: who it shuts out, the WCAG criteria, how to fix it and who usually does.
list_fixes: The AI-written fixes for a site, and where each one stands.
get_fix: One fix in full: what it changes, the live-page test result, and who did what.
get_site_health: One site at a glance: accessibility, speed, SEO and security scores, uptime, SSL expiry and broken links.
get_score_history: Scores over time, scan by scan.
what_changed: What is new, fixed or worse since the scan before.
diff_scans: Compare any two scans: what got fixed, what is new, what is still open.
list_alerts: What the daily monitor caught lately: a score drop, a new critical or serious issue, a regression.
get_scan_status: Whether a scan is running and when the last one finished.
get_plugin_status: Whether the WordPress plugin is installed, up to date and showing fixes on the site.
get_report_links: Links to hand a person: the site in Scanpoof, its evidence pack, VPAT, accessibility statement and Excel export.
get_conformance_report: Where the site stands against WCAG Level A and AA from its latest scan, with links to its VPAT and accessibility statement.
list_seo_suggestions: Page titles and search descriptions Scanpoof's AI has suggested, and where each one stands.

Check a page now (3)

check_url: Load one page of your site now and run every check a scan runs. Only on your own verified domains.
check_html: Check a piece of HTML before it ships, such as a component your agent is writing.
verify_fix: Reload the page an issue was on and say whether it is fixed. Much faster than a full rescan.

Act (10)

request_rescan: Start a fresh scan of a site.
request_fix_generation: Ask Scanpoof's AI to write and test fixes now.
propose_fix: Propose a fix of its own. It gets the same live-page test as any other.
approve_fix: Put a tested fix on the site. Only if you switched that on for the site.
approve_verified_fixes: Approve every fix that passed the live-page test, in one go. Same switch.
reject_fix: Reject a fix that hasn't gone live, with a reason.
disable_fix: Turn a live fix off. Always allowed, because it is the safe direction.
request_seo_suggestions: Ask Scanpoof's AI to suggest better page titles and search descriptions.
decide_seo_suggestion: Approve or reject one SEO suggestion. Approving needs the same switch; rejecting is always allowed.
approve_all_seo_suggestions: Approve every waiting SEO suggestion at once. Same switch.

Built so an agent can't overreach.

An AI agent can be wrong, or be talked into something by text on a web page. So the limits live in Scanpoof, not in the agent's instructions.

Each key covers all your sites or only the ones you pick
Each key can be read-only, or limited to a list of tools
Approving is off until you switch it on for a site
Text copied from your pages is labeled as untrusted in tool results
Revoke a key any time; Settings shows what each agent did

What an agent can't do

Approve a fix unless you switched that on for the site, or put live a fix that failed the test
Change your page's text, links, scripts or markup: a fix can only add or change a few attributes and styles, and an SEO suggestion only the page title and search description
Change anything at all with a read-only key, use a tool its key wasn't given, or touch a site its key wasn't given
Check pages outside your own verified domains
Add sites, change billing, invite people or see anyone else's account
Work at all unless the account has AI Fix
Go faster than fair use: 30 changes and 240 reads a minute per key

What teams use it for.

The morning check for an agency

Your agent goes through every client site, tells you which ones got worse since yesterday, asks for fixes on those and lists what needs a person.

An assistant your team already talks to

Connect Scanpoof to the AI assistant in your team chat. Anyone can ask “how is the site doing?” and get the score and the biggest problems. Changes still wait for someone allowed to approve them.

Checks before code ships

A coding agent can run check_html on a component it just wrote, or check_url on a staging page, and fix what it finds before anyone reviews the pull request.

A weekly report nobody has to write

The weekly-review prompt turns score history, fixes and new issues into a short note for the site owner. Your agent posts it where you already look.

Posting to chat, email or tickets is your agent's job, not Scanpoof's. Scanpoof gives it the facts and enforces the limits.

Ready-made prompts.

Pick one in your AI tool and the agent knows which Scanpoof tools to use, in which order, and when to stop and ask you.

Scan and fix a site: Finds the most urgent problems, gets fixes written and tested for the ones Scanpoof can fix, and lists what needs a person.
Fix one issue: Explains one issue, proposes a fix for it, and checks the live page once it is fixed.
Weekly review: A short report for the site owner: score trend, what got fixed, what is new, and what is waiting for a decision.

Connect it in a minute.

Make a key under Settings, AI agents (AI Fix is required), then add the server to your agent. The key works only with the MCP server; keys for CI or the WordPress plugin don't work there. The API docs have the full reference.

Claude Code

claude mcp add --transport http scanpoof https://scanpoof.com/api/mcp \
  --header "Authorization: Bearer adap_mcp_your_key_here"

Other MCP clients

{
  "mcpServers": {
    "scanpoof": {
      "url": "https://scanpoof.com/api/mcp",
      "headers": { "Authorization": "Bearer adap_mcp_your_key_here" }
    }
  }
}

Field names vary a little between tools. You need the address and the Authorization header.

Questions, answered.

Anything else? Email [email protected].

What is an MCP server?

MCP (Model Context Protocol) is an open standard that lets AI tools use other software. Scanpoof's MCP server gives an AI agent a fixed set of tools for your accessibility results. The agent decides what to ask; Scanpoof decides what it is allowed to do.

Which AI tools can connect?

Any tool that supports remote MCP servers over HTTPS and lets you set an Authorization header, such as Claude Code, Cursor or VS Code. Tools that only offer OAuth sign-in can't connect yet.

Can an agent change my site without me?

Only if you allow it. By default an agent can look at results, check pages, ask for new fixes, reject fixes and turn live ones off. It can approve a fix only when you switch that on for the site, and only a fix that passed its test on your live page. Every action is logged under the key's name.

Can I limit a key to some sites or some tools?

Yes. Each key can cover all your sites or only the ones you pick, can be read-only, and can be limited to a list of tools, for example look-only or page checks only. Give each agent its own key so you can revoke one without touching the others.

Does it work inside WordPress?

Yes. On WordPress 6.9 and newer, the Scanpoof plugin also registers five abilities with WordPress's Abilities API, so a WordPress-aware agent can read your summary, list fixes, reject one, turn one off, pause them all and ask for a scan from inside your site. They need a logged-in administrator, and approvals follow the same switch.

What about instructions hidden in a web page?

Text that comes from your pages is labeled as untrusted page content in the tool results, so a well-behaved agent treats it as data, not as orders. The limits on the key still apply whatever the agent decides.

Does it cost extra?

No. The MCP server comes with the AI Fix add-on, from $39 a month. There is no separate fee and no per-call charge. Fair-use limits apply: 30 changes and 240 reads a minute per key.

Does it make my site compliant?

No tool can promise that. Automated checks find a large share of common problems, and the agent helps you work through them faster, but some things still need a person to judge, such as whether a description says the right thing.

Ready to see where your site stands?

Scan a page free, or start a 14-day trial and monitor every site daily.

Start free trial Scan my site free