Skip to content

Yellowhead Digital agent door MCP server

digital.yellowhead/agent-dooryellowhead.digitalstreamable-httpv0.1.0Scanned

Read your Yellowhead Diagnostics, scans and action items in an AI assistant, and start Diagnostics.

21 tools: 1 change things and 20 only read.

Tools: 0 delete, 0 irreversible, 1 write, 20 read.
  • Reads private data · 1
  • Reads text others write · 6
  • Can send data out · 2
  • 1 finding

Tools listed without sign-in. Remote posture

With Anlyon

1 tool here changes things. Govern it.

  • Anlyon lists the tools and adopts none.
  • Each tool you adopt gets approvals and receipts.
  • Your agent never holds the key.
Govern this server →

A tool your agent calls directly is not routed through Anlyon and is not governed by it. Anlyon Radar only lists and labels, and these labels are the same either way.

What the Yellowhead Digital agent door MCP server can do

The Yellowhead Digital agent door MCP server is published by yellowhead.digital, a domain the MCP Registry verified. It has 21 tools. It can create or change data (1 tool: start_diagnostic), but cannot delete data. 20 tools only read.

1 tool reads private data (get_accessibility_scan); 6 tools bring in text other people can write (get_diagnostic, get_citation_scan, get_crawlability_scan and 3 more); 2 tools can send data out of your system (describe_yellowhead and start_diagnostic). On its own it has all three parts of a prompt-injection exfiltration path: private data, text others write, and a way out.

The fixed checks matched 1 finding (long-description). Each one comes with the text it matched.

Its remote endpoint lists its tools without sign-in, and its tools changed once since Anlyon Radar first saw them.

Last scanned Nov 20, 2026, version 0.1.0.

Tools

Grouped by what each tool does, most consequential first. Open one for its reasons and full description; descriptions are the publisher's own words.

21 tools

Changes things 1Claude Code: ask
  • start_diagnosticStarts a Diagnostic of a website for the signed-in person, in two calls.Reads text others writeCan send data out1 finding

    Why it is in this group

    • The name "start_diagnostic" says "start": a word for a change.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    • outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.

    Findings

    • Info

      The description is 2,707 characters long, over the 2,000 this check looks for.

      Matched text

      description ends: "...kspaces instead. Available on connections made through an AI assistant's sign-in."

      What this check looks for
    Description, as the server wrote it
    Starts a Diagnostic of a website for the signed-in person, in two calls. The first call returns a preview and starts nothing: the website, the business name, the ten 1-to-10 self-ratings (which were answered, carried over or entered as 5 for "not sure"), the monthly tools budget, what the run is recorded as and where it is filed, which of the account's included monthly Diagnostics it uses, and a confirmation_token. A second call with that confirmation_token starts it; any change needs a new preview. When the website has versions for several countries or languages (acme.com/uk, acme.com/au), the preview lists them in site_versions, and which one is checked is the person's choice; a version is checked through a new preview with its url. A site is a domain, a subdomain or a country or language folder (acme.com/uk), and workspaces are matched by site: when a domain has workspaces for some of its folders and the address is in none of them (host_split), the run is not filed in any of them automatically, and the preview or needs_client lists them with the site each one is for. It uses only the account's included Diagnostics: when none are left the result is quota_exhausted with the reset date and the plans page, and nothing is charged. A run takes 5 to 10 minutes and a completion email also arrives; get_run_status reports progress. earlier_report_id fills in the website, name, ratings and budget from an earlier report, for a re-check; if that run did not complete and the same person started it, starting it again once for the same website and name does not use another Diagnostic (the preview says whether it applies). file_as says what the run is for, with the same five choices as the app: my_business (the account's own website), client, competitor, research or prospect. A competitor or research run of a website no workspace holds is filed in no workspace and creates none; a website or business name a client or brand workspace holds is filed in that workspace, as in the app. On an account with workspaces (clients, prospects or brands), client_id names the workspace, and needs_client lists the candidates when where the run is filed is not clear. For an owner or admin of an Agency-plan team account, file_as prospect on a website not yet in Prospects creates a new prospect workspace for it (named by prospect_name, else the business name); the preview says so and nothing is created before the confirmation. The same person may ask for a new client workspace for a site no workspace holds: file_as client with client_name (the new workspace's name); without client_name a miss lists the client workspaces instead. Available on connections made through an AI assistant's sign-in.

    Annotations the server declares

    • title: Start a free Diagnostic
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
Only reads 20Show 20 read-only tools
  • describe_yellowheadDescribes Yellowhead Digital, with no sign-in needed: what it is; the free Diagnostic (the ten pillars it checks, how long it takes, what the Free plan includes, and that an accoun…Reads text others writeCan send data outNo findings

    Why it is in this group

    • The name "describe_yellowhead" says "describe": a word for reading.

    What it can reach

    • untrusted-inputThe description's first sentence says "describe" and "website": it reads web pages, which anyone can write.
    • untrusted-inputThe input takes a URL (input.url), and the tool reads it: it reads web pages, which anyone can write.
    • outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.

    Claude Code: ask, because it is tagged outbound.

    Description, as the server wrote it
    Describes Yellowhead Digital, with no sign-in needed: what it is; the free Diagnostic (the ten pillars it checks, how long it takes, what the Free plan includes, and that an account is needed to run one); every plan with its price and monthly Diagnostics; the paid scans and human services with their prices and public pages; and a link that opens the Diagnostic form with a website and business name filled in. Fits questions about what Yellowhead does or costs, whether it checks something, or how to get started, including before any account exists. The optional url and business_name only fill in that link: nothing is scanned, stored or looked up.

    Annotations the server declares

    • title: About Yellowhead and the free Diagnostic
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_accessibility_scanReturns one Accessibility Scan by id: its website, sitemap, status, dates and page counts, and once completed its results: the run summary (pages measured on both viewports, templa…Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "get_accessibility_scan" says "get": a word for reading.

    What it can reach

    • private-dataThe description's first sentence says "return" and "note": it reads notes.
    • untrusted-inputThe description's first sentence says "return" and "website": it reads web pages, which anyone can write.
    Description, as the server wrote it
    Returns one Accessibility Scan by id: its website, sitemap, status, dates and page counts, and once completed its results: the run summary (pages measured on both viewports, templates sampled, and the note on what automated checks can and cannot show), the per-template roll-up of failing checks (the 20 largest templates and up to 25 of the most widespread failing checks each, with the totals), and on a re-scan the fix verification against the earlier scan as counts (fixed, still failing or not confirmed). results is null until the scan completes; on a completed scan whose stored results could not be read, results_unavailable is true and the results are in the Yellowhead app. Automated checks cover only part of WCAG, so a page with no failing check is not thereby shown to meet it. The per-page results are in the Yellowhead app.

    Annotations the server declares

    • title: Get an Accessibility Scan
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_action_itemReturns one action item by id with its full detail: why it matters (rationale), what to do (description and remediation guidance), the tools involved, its status and dates, and the…No findings

    Why it is in this group

    • The name "get_action_item" says "get": a word for reading.
    Description, as the server wrote it
    Returns one action item by id with its full detail: why it matters (rationale), what to do (description and remediation guidance), the tools involved, its status and dates, and the source it came from (source_type and source_id: a Diagnostic, or a Citation, Crawlability or Accessibility Scan). Fits explaining or planning one fix. Fixes happen on the business's own systems with its own access; an item becomes verified only when a later Diagnostic sees the fix from outside. Changes to an action item are made in the Yellowhead app, or through a connection an account admin set up there.

    Annotations the server declares

    • title: Get an action item
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_citation_scanReturns one Citation Scan by id: its website, category, country, status and dates, the named competitors (competitor_domains), and once completed its results: the ranking of every …Reads text others writeNo findings

    Why it is in this group

    • The name "get_citation_scan" says "get": a word for reading.

    What it can reach

    • untrusted-inputThe description's first sentence says "return" and "website": it reads web pages, which anyone can write.
    Description, as the server wrote it
    Returns one Citation Scan by id: its website, category, country, status and dates, the named competitors (competitor_domains), and once completed its results: the ranking of every domain the answers cited with each one's share of answered cells, the coverage (how many answer cells were filled), each answer cell's question, engine, whether the scanned website was cited and which domains were, and where the scan asked them, the head-to-head questions against named competitors. results is null until the scan completes; on a completed scan whose stored results could not be read, results_unavailable is true and the results are in the Yellowhead app. The citations are what the engines answered when asked; they can differ between runs and are not on their own evidence of why a site was or was not cited. The cited page addresses are in the Yellowhead app.

    Annotations the server declares

    • title: Get a Citation Scan
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_crawlability_scanReturns one Crawlability Scan by id: its website, sitemap, status, dates and page counts (imported, approved, measured, not measured, dead, stale or missing last-modified dates, du…Reads text others writeNo findings

    Why it is in this group

    • The name "get_crawlability_scan" says "get": a word for reading.

    What it can reach

    • untrusted-inputThe description's first sentence says "return" and "website": it reads web pages, which anyone can write.
    Description, as the server wrote it
    Returns one Crawlability Scan by id: its website, sitemap, status, dates and page counts (imported, approved, measured, not measured, dead, stale or missing last-modified dates, duplicates, linked but not in the sitemap), and once completed its run summary (results): whether robots.txt and llms.txt were read, the per-crawler roll-up of what robots.txt allows, how pages answered AI crawler identities compared with a browser, pages linked from the site that the sitemap does not list (up to 25 of each list, with the totals), and why a run stopped early if it did. results is null until the scan completes; on a completed scan whose stored results could not be read, results_unavailable is true and the results are in the Yellowhead app. A robots.txt that could not be read is reported as unknown, not as absent. The per-URL results are in the Yellowhead app.

    Annotations the server declares

    • title: Get a Crawlability Scan
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_diagnosticReturns one Diagnostic by id: its status, business name, website, client_id, MIDAS score and created date, plus report_id once it has completed.Reads text others writeNo findings

    Why it is in this group

    • The name "get_diagnostic" says "get": a word for reading.

    What it can reach

    • untrusted-inputThe description's first sentence says "return" and "website": it reads web pages, which anyone can write.
    Description, as the server wrote it
    Returns one Diagnostic by id: its status, business name, website, client_id, MIDAS score and created date, plus report_id once it has completed. Fits checking whether a particular run has finished; the findings themselves come from get_report.

    Annotations the server declares

    • title: Get a Diagnostic
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_infrastructure_auditReturns one published Infrastructure Audit by id, as published to the account.No findings

    Why it is in this group

    • The name "get_infrastructure_audit" says "get": a word for reading.
    Description, as the server wrote it
    Returns one published Infrastructure Audit by id, as published to the account. Fits working from its findings and recommendations.

    Annotations the server declares

    • title: Get an Infrastructure Audit
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_portfolio_changesLists the client workspaces whose MIDAS score moved since a date (default: the last 30 days), largest move first.No findings

    Why it is in this group

    • The name "get_portfolio_changes" says "get": a word for reading.
    Description, as the server wrote it
    Lists the client workspaces whose MIDAS score moved since a date (default: the last 30 days), largest move first. A client appears when its most recent scored Diagnostic is on or after that date and its MIDAS changed, against the scored Diagnostic before it, by more than the usual run-to-run variation; smaller moves are left out, because scores for an unchanged site can move a little between runs. Each item has the client_id and client name, the previous and current MIDAS and the change, both report_ids with their dates for citing (report_id, previous_report_id and previous_report_date are null when that pair cannot be confirmed in this account), and the open items of critical or high priority (only work the account owes: items from Diagnostics classified as a competitor, research or a prospect are not counted). direction narrows the list to declines or improvements; client_id narrows it to one client. An empty list means no client moved by more than that in the window, not that nothing changed. truncated is true when more clients moved by more than that in the window than one call examines (the most recent are examined), so the list may be incomplete. Archived workspaces are not included. Included with team accounts on the Agency plans, Business Enterprise and custom plans. Fits "which of my clients got worse this month" and "who improved since our last review".

    Annotations the server declares

    • title: Get Portfolio changes
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_reportReturns the report of a completed Diagnostic by id: the ten pillar scores as the person rated them (side_a_scores) and as detected (side_b_scores), the MIDAS score, executive summa…No findings

    Why it is in this group

    • The name "get_report" says "get": a word for reading.
    Description, as the server wrote it
    Returns the report of a completed Diagnostic by id: the ten pillar scores as the person rated them (side_a_scores) and as detected (side_b_scores), the MIDAS score, executive summary, blind spots, budget and investment reads, business profile, detected tech stack, brand visibility and sentiment, published apps, and a counts-only summary of the AI connectors found. Internal signals are not included. Some detail depends on the account's plan, as the app shows it: on the Free plan the report carries one blind spot (with that pillar's sub-scores only), the brand visibility and sentiment scores without their findings, the stated and estimated tool spend (on older reports, the estimated total) without the per-pillar breakdown, and no market context; the advertising sub-scores also need Starter or above; the investment roadmap (recommended_investment) and published-app identifiers and links need Pro or above (below Pro, published_apps lists only the platforms). A Diagnostic that has not finished returns report_not_ready. Scores for an unchanged site can move a little between runs, so a small change is not on its own evidence that something changed. The findings are what can be observed from outside.

    Annotations the server declares

    • title: Get a Diagnostic report
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_run_statusReports whether a Diagnostic is queued, running, completed or failed, by its id (diagnostic_id, as start_diagnostic and list_diagnostics give it), with when it started and how many…No findings

    Why it is in this group

    • The name "get_run_status" says "get": a word for reading.
    Description, as the server wrote it
    Reports whether a Diagnostic is queued, running, completed or failed, by its id (diagnostic_id, as start_diagnostic and list_diagnostics give it), with when it started and how many minutes have passed. A Diagnostic usually takes 5 to 10 minutes; one taking longer is reported as running with slow set, because it may still finish, and as failed only once it has recorded a failure or 30 minutes have passed. A failed result says whether starting it again from this connection uses another of the month's Diagnostics. Fits "is it ready yet?" at any point in the conversation.

    Annotations the server declares

    • title: Get a Diagnostic run's status
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_strategic_assessmentReturns one published Strategic Assessment by id: the deliverable written for the business after its live session.No findings

    Why it is in this group

    • The name "get_strategic_assessment" says "get": a word for reading.
    Description, as the server wrote it
    Returns one published Strategic Assessment by id: the deliverable written for the business after its live session. Fits working from its priorities and recommendations.

    Annotations the server declares

    • title: Get a Strategic Assessment
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_usageReturns this month's Diagnostic allowance for the connected account: how many included Diagnostics are used (used), the monthly allowance (limit), how many remain (remaining), the …No findings

    Why it is in this group

    • The name "get_usage" says "get": a word for reading.
    Description, as the server wrote it
    Returns this month's Diagnostic allowance for the connected account: how many included Diagnostics are used (used), the monthly allowance (limit), how many remain (remaining), the plan, whose allowance a run draws on (pool), and the date the allowance resets (resets_on). On the Free plan it carries a note on the first-month allowance. With client_id, it also gives that client workspace's own monthly limit (per_client). Fits questions about how many Diagnostics are left this month and when the allowance resets.

    Annotations the server declares

    • title: Get Diagnostic usage
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_accessibility_scansLists the Accessibility Scans in the connected account, newest first.No findings

    Why it is in this group

    • The name "list_accessibility_scans" says "list": a word for reading.
    Description, as the server wrote it
    Lists the Accessibility Scans in the connected account, newest first. An Accessibility Scan measures a sample of the website's pages against automated WCAG 2.x A and AA checks, grouped by page template. Each item has its id, the website (target_domain), status (draft, importing, awaiting_review, pending_payment, paid, running, completed, failed or cancelled; the first three are the free sitemap import and page review), whether it is a re-scan of an earlier scan's pages (sample_mode same_as_prior) or a first scan (fresh), the approved and measured page counts, the number of page templates, client_id, diagnostic_report_id, and created and completed dates. Filters: client_id and status. An empty page means none were returned for these filters, not that the account has none or that Yellowhead does not offer the scan.

    Annotations the server declares

    • title: List Accessibility Scans
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_action_itemsLists the action items in the connected account, newest first.No findings

    Why it is in this group

    • The name "list_action_items" says "list": a word for reading.
    Description, as the server wrote it
    Lists the action items in the connected account, newest first. Each has its id, title, description, rationale, status, priority, pillar, phase, estimated cost range, suggested tools, remediation guidance, the source it came from (source_type and source_id: a Diagnostic, which get_report reads, or a Citation, Crawlability or Accessibility Scan, which get_citation_scan, get_crawlability_scan or get_accessibility_scan reads), client_id, assignee and due date. Statuses: suggested, queued, in_progress, completed (the person doing it says it is done), confirmed (a second person signed off), verified (a later Diagnostic saw the fix from outside) and dismissed. Filters: client_id, source_id and status. An empty page means none were returned for these filters. Fits "what should we fix first" and progress reviews. Changes to an action item are made in the Yellowhead app, or through a connection an account admin set up there.

    Annotations the server declares

    • title: List action items
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_citation_scansLists the Citation Scans in the connected account, newest first.No findings

    Why it is in this group

    • The name "list_citation_scans" says "list": a word for reading.
    Description, as the server wrote it
    Lists the Citation Scans in the connected account, newest first. A Citation Scan asks a fixed set of buyer questions in one category and country across several AI answer engines and records which websites the answers cite. Each item has its id, the website scanned (target_domain), category, country, status (pending_payment, paid, running, completed, failed or cancelled), client_id, the Diagnostic it started from (diagnostic_report_id), created and completed dates, and once completed the share of answered cells that cited the website (target_share, 0 to 1) and how many answer cells were filled (populated_cells). Filters: client_id and status. An empty page means none were returned for these filters, not that the account has none or that Yellowhead does not offer the scan.

    Annotations the server declares

    • title: List Citation Scans
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_crawlability_scansLists the Crawlability Scans in the connected account, newest first.No findings

    Why it is in this group

    • The name "list_crawlability_scans" says "list": a word for reading.
    Description, as the server wrote it
    Lists the Crawlability Scans in the connected account, newest first. A Crawlability Scan checks, URL by URL from the website's sitemap, whether AI crawlers are allowed in by robots.txt and whether pages answer AI crawler identities the way they answer a browser. Each item has its id, the website (target_domain), status (draft, importing, awaiting_review, pending_payment, paid, running, completed, failed or cancelled; the first three are the free sitemap import and page review), whether it is a re-scan of an earlier scan's pages (sample_mode same_as_prior) or a first scan (fresh), the approved and measured page counts, client_id, diagnostic_report_id, and created and completed dates. Filters: client_id and status. An empty page means none were returned for these filters, not that the account has none or that Yellowhead does not offer the scan.

    Annotations the server declares

    • title: List Crawlability Scans
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_diagnosticsLists the Diagnostics in the connected Yellowhead account, newest first.No findings

    Why it is in this group

    • The name "list_diagnostics" says "list": a word for reading.
    Description, as the server wrote it
    Lists the Diagnostics in the connected Yellowhead account, newest first. Each item has its id, business name, website, status (queued, processing, completed or failed; a run with no result after 15 minutes is reported as failed), client_id, MIDAS score once complete, and created date. Diagnostics are visible across the whole account, not only to the person who ran them. Filters: client_id (one client workspace) and since (created on or after a date and time). An empty page means none were returned for these filters, not that none exist. Fits questions like "what Diagnostics do we have" and "is my Diagnostic ready", and finding two runs to compare.

    Annotations the server declares

    • title: List Diagnostics
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_infrastructure_auditsLists the published Infrastructure Audits in the connected account, newest first, with their ids, the Diagnostic each followed, client_id and dates.No findings

    Why it is in this group

    • The name "list_infrastructure_audits" says "list": a word for reading.
    Description, as the server wrote it
    Lists the published Infrastructure Audits in the connected account, newest first, with their ids, the Diagnostic each followed, client_id and dates. Only published audits appear: one that is in progress appears once it is published. Filters: client_id and diagnostic_report_id.

    Annotations the server declares

    • title: List Infrastructure Audits
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_portfolioLists the client workspaces of the connected team account (brand workspaces on Business Enterprise), one page at a time, with the total so an answer can say how many are shown out …No findings

    Why it is in this group

    • The name "list_portfolio" says "list": a word for reading.
    Description, as the server wrote it
    Lists the client workspaces of the connected team account (brand workspaces on Business Enterprise), one page at a time, with the total so an answer can say how many are shown out of how many. Each row has the client_id, client name and website, the latest Diagnostic (report_id and date; it may still be running or may have failed) and the days since it, the MIDAS score with its tier, its change from the previous scored Diagnostic and whether that change is larger than the usual run-to-run variation, open_action_items (the Portfolio page's own count: items not done and not dismissed, so suggested, queued and in progress), open_high_priority_items (the open items that are critical or high), and the next scheduled Diagnostic. Both item counts cover only work the account owes: items from Diagnostics classified as a competitor, research or a prospect are not counted (list_action_items still lists them). Sorts: name, largest_decline, largest_improvement, lowest_midas, most_open_high_priority, most_open_items, longest_since_last_diagnostic, most_recent_diagnostic and next_scheduled; in the score and date sorts, a client with no Diagnostic or no score comes last. Filters: search (name or website), status, MIDAS tier, client_id, pillar (counts only that pillar's critical and high items) and min_open_high_priority_items. An empty page means no workspace matched these filters, not that the account has none. Included with team accounts on the Agency plans, Business Enterprise and custom plans. Fits "which clients need attention", ranking clients, and finding a client's client_id by name for list_action_items or list_diagnostics.

    Annotations the server declares

    • title: List the Portfolio
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_strategic_assessmentsLists the published Strategic Assessments in the connected account, newest first, with their ids, the Diagnostic each was based on, client_id and dates.No findings

    Why it is in this group

    • The name "list_strategic_assessments" says "list": a word for reading.
    Description, as the server wrote it
    Lists the published Strategic Assessments in the connected account, newest first, with their ids, the Diagnostic each was based on, client_id and dates. Only published assessments appear: one that is booked or still being written appears once it is published, so an empty page does not mean none was ordered. Filters: client_id and diagnostic_report_id. A Strategic Assessment is a written walkthrough of a Diagnostic's findings, prepared with the business: which matter and what to fix first.

    Annotations the server declares

    • title: List Strategic Assessments
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false

This is not a safety rating, and no finding is not a safety guarantee. The groups and findings come from fixed checks of the tool definitions the server returned on the date shown, not from its code or what it does when called. How the checks work.

Exfil surface

Which tools read private data, which read text someone else can write, and which can send data out. One of each is a path a prompt injection can use.

12 paths in this server alone from private data, through text someone else wrote, to a way out. Denying 1 private-data tool breaks every one.

Reads private data1
Reads files, mail, messages, databases, secrets, calendars or repositories.
Reads text others write6
Brings in text someone else can write: web pages, inbound email, messages, issues and comments.
Can send data out2
Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.
deny to break every path
{
  "permissions": {
    "deny": [
      "mcp__agent-door__get_accessibility_scan"
    ]
  }
}

Other servers in the same agent add their own tools to these paths. Audit your whole setup to see them together.

Remote posture

What anyone on the internet sees at the remote endpoint: an anonymous initialize and tools/list, and the public OAuth metadata.

Tools listed without sign-in

Looked at Nov 21, 2026, 19:22 UTC, unchanged since Nov 14, 2026. No credentials were sent and no tool was called. What is looked at.

Reachable
yes
HTTP status of an anonymous initialize
200
tools/list served without credentials
yes
Tools that change things, listed without credentials
1
Protocol version
2025-06-18
WWW-Authenticate on 401
not observed
resource_metadata in WWW-Authenticate
not observed
RFC 9728 protected resource metadata
https://api.yellowhead.digital/.well-known/oauth-protected-resource/mcp
Authorization servers
https://api.yellowhead.digital
Legacy HTTP+SSE transport
no
Dynamic client registration
yes
Client ID metadata documents
yes
  • WarnThe whole server

    The endpoint served tools/list without credentials, and the list has 1 tool that changes things (1 write). Whether a call also works without credentials was not tested: tools are never called.

    Matched text

    https://api.yellowhead.digital/mcp: tools/list returned 21 tools to an anonymous client, including start_diagnostic (write)

    What this check looks for

Posture changes

  1. Nov 14, 2026
    • Tools that change things, listed without credentials: 4 to 1

Findings

Matches of fixed checks, with the text they matched.

1 info
  • The description is 2,707 characters long, over the 2,000 this check looks for.

    Matched text

    description ends: "...kspaces instead. Available on connections made through an AI assistant's sign-in."

    What this check looks for

History

Changes to its tool definitions and to its remote endpoint, newest first.

  1. Tools4 removed, 7 changed.−4 removed~7 changedView diff
  2. PostureRemote posture changed. Tools that change things, listed without credentials: 4 -> 1.View diff
    • Tools that change things, listed without credentials: 4 to 1

Install

Pinned to version 0.1.0. Values in angle brackets are yours to fill in.

claude mcp add --transport http agent-door https://api.yellowhead.digital/mcp
claude mcp add --transport http agent-door https://api.yellowhead.digital/mcp

Run in a terminal.

Claude Code permissions

Allow the reads, ask before a change or a way out, deny the deletes. Paste into .claude/settings.json.

.claude/settings.json
{
  "permissions": {
    "allow": [
      "mcp__agent-door__list_diagnostics",
      "mcp__agent-door__get_diagnostic",
      "mcp__agent-door__get_report",
      "mcp__agent-door__list_action_items",
      "mcp__agent-door__get_action_item",
      "mcp__agent-door__get_usage",
      "mcp__agent-door__list_strategic_assessments",
      "mcp__agent-door__get_strategic_assessment",
      "mcp__agent-door__list_infrastructure_audits",
      "mcp__agent-door__get_infrastructure_audit",
      "mcp__agent-door__list_citation_scans",
      "mcp__agent-door__get_citation_scan",
      "mcp__agent-door__list_crawlability_scans",
      "mcp__agent-door__get_crawlability_scan",
      "mcp__agent-door__list_accessibility_scans",
      "mcp__agent-door__get_accessibility_scan",
      "mcp__agent-door__list_portfolio",
      "mcp__agent-door__get_portfolio_changes",
      "mcp__agent-door__get_run_status"
    ],
    "ask": [
      "mcp__agent-door__describe_yellowhead",
      "mcp__agent-door__start_diagnostic"
    ],
    "deny": []
  }
}

Tools that only read are allowed, tools that change things ask first, tools that delete or cancel are denied. A read that can send data out (it takes a URL, say) asks first too. It assumes the server is installed as agent-door. Review it before you use it.

Capability tags behind 6 rules
  • mcp__agent-door__get_diagnostic untrusted-input
  • mcp__agent-door__get_citation_scan untrusted-input
  • mcp__agent-door__get_crawlability_scan untrusted-input
  • mcp__agent-door__get_accessibility_scan private-data, untrusted-input
  • mcp__agent-door__describe_yellowhead untrusted-input, outbound
  • mcp__agent-door__start_diagnostic untrusted-input, outbound

Pin

Keep the decision: fail CI when the tools change, follow every change, or show a badge.

Pin in CI

Fail a build when the tools you reviewed are not the tools the server serves today.

curl -s "https://radar.anlyon.com/api/v1/servers/digital.yellowhead/agent-door/verify?hash=2017776f63f8f05eeae544cbfead08bc682d058a6c62318506c0a9f7d08bd010" | jq -e '.changed == false'

Pinned hash 2017776f63f8, from Nov 20, 2026, 22:17 UTC.

Watch

Get told when its tools change: a new tool, a new description, a new class.

https://radar.anlyon.com/servers/digital.yellowhead/agent-door/changes.xml

An Atom feed. Works in any feed reader or Slack's RSS app.

Badge

Anlyon Radar badge for digital.yellowhead/agent-door
[![Anlyon Radar](https://radar.anlyon.com/badge/digital.yellowhead/agent-door.svg)](https://radar.anlyon.com/servers/digital.yellowhead/agent-door)

Source: MCP Registry. Synced Nov 6, 2026, 20:17 UTC.

Yellowhead Digital agent door MCP server: questions and answers

Answered from the scan above, not from the publisher's description.

Is the Yellowhead Digital agent door MCP server safe?

No scan can promise that. What its definitions show: 21 tools, 1 of which can change, send or delete something, and 1 finding from fixed checks. Allow the reads, ask before each change, and deny the deletes; the permissions block on this page does that.

Can the Yellowhead Digital agent door MCP server delete data?

No tool is classed as destructive: going by their definitions, none of them deletes, cancels or revokes anything.

Can the Yellowhead Digital agent door MCP server do things that cannot be undone?

No tool is classed as irreversible: none of them sends, pays, publishes or merges, going by their definitions.

What tools does the Yellowhead Digital agent door MCP server have?

21 tools: 1 write, 20 read. start_diagnostic, list_diagnostics, get_diagnostic, get_report, list_action_items, get_action_item, get_usage, list_strategic_assessments and 13 more.

Does the Yellowhead Digital agent door MCP server need sign-in or OAuth?

Its remote endpoint listed its tools to an anonymous client, without sign-in.

Does the Yellowhead Digital agent door MCP server send data to outside services?

2 tools can carry data out (it sends, posts or publishes, or takes a recipient, channel or URL): describe_yellowhead and start_diagnostic.

Has the Yellowhead Digital agent door MCP server changed its tools?

Yes, once since Anlyon Radar started tracking it, most recently on Nov 20, 2026. Every change is on this page and in its Atom feed.