MCP server

Use Citehound from inside Claude.

Scan a site, get the fix, read the playbook and the research, without leaving the conversation. Free, no signup.

Connect it See what it can do
Example
You ask

Scan example.com and tell me what to fix first

citehound · scan_site example.com
Citehound returns
AI readiness scan — example.com Score: 29/100 Discoverability: 26/40 Technical foundation: 3/20 Content & trust: 0/40 That’s 44 points below the average of the 219 sites we’ve scanned (73/100). AI crawler access (10 tracked): GPTBot — Open ChatGPT-User — Open

Real output from the tool, October 8, 2026. Excerpt.

Ask

Ask in plain language.

“Scan example.com and tell me what to fix first”

scan_site

“Compare these two sites”

compare_sites

“Write me an llms.txt for my business”

generate_llms_txt

“Which questions should I test for my category?”

get_citation_prompts

Tools

What’s inside.

The server has thirteen tools in four groups. Every one is read-only.

Scan

Read a live site the way an AI crawler would.

scan_site

Score any site out of 100 for AI readiness, with every failed check and its fix.

Inputs and example

Inputs

Inputs for scan_site
InputWhat it takes
domainrequiredA bare domain, e.g. "example.com". No scheme (https://) and no path — the scan always targets the domain root.

Example

You ask: “Scan getcitehound.com and tell me what to fix first”

Real output, first 12 of 25 lines
AI readiness scan — getcitehound.com

Score: 87/100
  Discoverability:      27/40
  Technical foundation: 20/20
  Content & trust:      40/40

That’s 14 points above the average of the 219 sites we’ve scanned (73/100).

AI crawler access (10 tracked):
  GPTBot — Limited (Disallow: /admin/ (+1 more) · bot-specific rule)
  ChatGPT-User — Limited (Disallow: /admin/ (+1 more) · bot-specific rule)

Captured October 8, 2026.

compare_sites

Put two sites side by side and see every check where they differ.

Inputs and example

Inputs

Inputs for compare_sites
InputWhat it takes
domain_arequiredThe first bare domain, e.g. "example.com".
domain_brequiredThe second bare domain, e.g. "competitor.com". Must differ from domain_a.

Fix

Generate the files that close the gaps.

generate_schema

Get paste-ready JSON-LD for your site, with brackets where only you know the facts.

Inputs and example

Inputs

Inputs for generate_schema
InputWhat it takes
domainrequiredA bare domain, e.g. "example.com".
typerequired"organization" returns Organization + WebSite schema as an @graph; the others (faqpage, article, product, localbusiness) each return one schema of that @type.

Example

You ask: “Write Organization JSON-LD for getcitehound.com”

Real output, first 12 of 19 lines
Organization and WebSite JSON-LD for getcitehound.com, ready to paste inside a <script type="application/ld+json"> tag in the page’s <head>:

{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "Organization",
      "name": "[Your company name]",
      "url": "https://getcitehound.com",
      "description": "Scan a site for AI visibility, free: 16 checks, 10 AI crawlers, live data. Then fix it with vertical-specific GEO & AEO playbooks. No email required."
    },
    {

Captured October 8, 2026.

generate_robots_txt

Allow or block named AI crawlers with a robots.txt you can paste as it is.

Inputs and example

Inputs

Inputs for generate_robots_txt
InputWhat it takes
allowoptionalCrawler names to explicitly allow, e.g. ["GPTBot", "ClaudeBot"]. Must be names from the tracked list.
blockoptionalCrawler names to explicitly block. Must be names from the tracked list, and must not overlap with "allow".
sitemap_urloptionalOptional absolute sitemap URL to add as a Sitemap: line, e.g. "https://example.com/sitemap.xml".

generate_llms_txt

Write an llms.txt that tells AI systems what your site is and where to start.

Inputs and example

Inputs

Inputs for generate_llms_txt
InputWhat it takes
namerequiredThe site or product name for the llms.txt header.
descriptionrequiredA one-sentence summary of what the site is.
key_pagesoptionalPages worth pointing an AI system to, most important first.

Learn

Read the data, the method and the playbooks behind the score.

get_playbook

Read the playbook for your vertical: the shift, three strategies and the pitfalls to avoid.

Inputs and example

Inputs

Inputs for get_playbook
InputWhat it takes
verticaloptionalOptional. A vertical slug. B2B SaaS: crm, martech, hrtech, fintech, cybersecurity, devtools. Consumer brands: ecommerce, consumerapps, hospitality, marketplaces. Professionals: health, localservices, realestate, legal. Omit to list them all.

get_benchmark

See where your score sits against the sites in your category.

Inputs and example

Inputs

Inputs for get_benchmark
InputWhat it takes
categoryoptionalOptional. One of: crm, cybersecurity, devtools, dtc-brands, consumer-apps, hospitality. Omit to get every category.

Example

You ask: “How do CRM sites score on AI readiness?”

Real output, first 12 of 38 lines
AI readiness benchmark: CRM software (B2B SaaS)
Scanned 31 platforms as of 2026-07-29 (8 more could not be reached).

Score — average 78/100, median 80, range 53–91.
  Discoverability:      27/40
  Technical foundation: 18/20
  Content & trust:      33/40

Crawler access: 10% of sites block at least one AI crawler.
By crawler, in number of sites (31 scanned):
  GPTBot           blocked 1, limited 24, open 6
  ChatGPT-User     blocked 1, limited 24, open 6

Captured October 8, 2026.

list_ai_crawlers

See the ten AI crawlers we track and what each one is for.

Inputs and example

Inputs

No inputs.

get_methodology

See how the 100 points are built, check by check, and what the score cannot show.

Inputs and example

Inputs

Inputs for get_methodology
InputWhat it takes
checkoptionalOptional. A check label exactly as scan_site reports it, e.g. "Page title" or "AI crawler access". Case does not matter. Omit for the whole methodology.

Example

You ask: “How is the Citehound score built?”

Real output, first 12 of 32 lines
How the Citehound AI readiness score is built: 100 points across 3 pillars and 16 checks. Every point value below is read from the scanner that runs.

Discoverability: 40 points. Can AI crawlers reach the site at all. Access is the precondition for every other signal.
  - robots.txt present (4 pts): Whether https://<domain>/robots.txt can be read. A real 404 counts as default-open access; a failed fetch stops the scan. Why it matters: AI crawlers read your access policy from this file.
  - llms.txt present (4 pts): Whether https://<domain>/llms.txt exists and is a usable text file. Why it matters: An emerging standard that gives AI systems a curated map of your content.
  - Sitemap declared (6 pts): Whether robots.txt has a Sitemap: line, or /sitemap.xml exists as a urlset or sitemapindex. Why it matters: Sitemaps let crawlers discover your content completely.
  - AI crawler access (26 pts): For each of the 10 tracked AI crawlers: open (no rule applies), limited (at least one Disallow rule applies, often an ordinary path) or blocked (the whole site is disallowed). Open earns full credit, limited half, blocked none, scaled to the 26 points. Why it matters: Every blocked crawler removes you from that platform’s answers.

Technical foundation: 20 points. Baseline machine-readability hygiene.
  - Canonical tag (4 pts): Whether the homepage declares a rel=canonical link. Why it matters: Tells machines the definitive URL and prevents duplicate-content ambiguity.
  - html lang attribute (3 pts): Whether the <html> element declares a language. Why it matters: Lets AI systems classify your content’s language correctly.
  - Page title (3 pts): Whether the <title> is 10 to 70 characters. Why it matters: Answer engines frequently use the title as your source label.

Captured October 8, 2026.

get_research

Read our published research, with the figures and how we got them.

Inputs and example

Inputs

Inputs for get_research
InputWhat it takes
slugoptionalOptional. A report slug from the list this tool returns with no argument. Omit to list the reports.

get_sample_report

Open a full-site report on our own site, before and after we fixed what it found.

Inputs and example

Inputs

No inputs.

Test

Check for yourself whether an assistant names a brand.

get_citation_prompts

Get the buying-intent questions for your category to try in your own assistant.

Inputs and example

Inputs

Inputs for get_citation_prompts
InputWhat it takes
verticalrequiredA vertical slug. Valid: consumerapps, crm, cybersecurity, devtools, ecommerce, fintech, health, hospitality, hrtech, legal, localservices, marketplaces, martech, realestate. Question sets exist for: consumerapps, crm, cybersecurity, devtools, ecommerce, fintech, health, hospitality, hrtech, legal, localservices, marketplaces, martech, realestate.

get_citation_sample

See what a citation run looks like: a real, anonymized sample with the spread shown.

Inputs and example

Inputs

No inputs.

Example

You ask: “Show me a sample citation result”

Real output, first 12 of 38 lines
Citation tracking sample: a CRM brand
Model: gemini-3.5-flash-lite. Date: October 2, 2026. No web search was enabled.
Each of 18 questions was asked 5 times (5 tries per question), each in its own conversation.

Overall, the brand was named in 49% of answers. Taking one try of every question as a sweep, the rate ranged from 39% to 56% across 5 sweeps. The average hides that spread.

Named in every try (6):
  - Which CRM should a 10-person startup use that won't need a consultant to set up?
  - What are the best CRMs that integrate natively with Slack?
  - Which CRM is easiest for a non-technical sales team to adopt?
  - What CRM should an agency use to manage clients and deals together?
  - Which CRMs integrate well with Google Workspace and Gmail?

Captured October 8, 2026.

Prompts

Ready-made workflows.

The server has three prompts that run several tools in order. Clients list them next to the tools, often as slash commands.

geo_audit

A full audit in one go: scan, benchmark, playbook, fixes and an ordered plan.

Callsscan_sitegenerate_schemagenerate_robots_txtgenerate_llms_txtget_playbookget_benchmarklist_ai_crawlersget_methodology

commerce_readiness

Check a store the way a shopping agent would, then get the fixes.

Callsscan_sitegenerate_schemagenerate_llms_txtget_playbook

citation_questions

Get the question set and the steps for testing whether an assistant names your brand.

Callsget_citation_promptsget_citation_sample

Connect

Connect it in a minute.

Add this address to your client. No key, no account. Connected before the rename? Remove the old connector and add this one again: tool prefixes change with the server name.

Claude
  1. Open Claude and go to Customize, then Connectors.
  2. Choose Add, then Add custom connector.
  3. Name it Citehound and paste the server address from above.
  4. Save it. No authentication is needed.
  5. Ask: “Scan example.com and tell me what to fix first.”

Custom connectors depend on your plan. Your client’s help has the current steps.

Claude Code
  1. Run this once, from any directory.
  2. Start a session and ask for a scan.
Terminal 
claude mcp add --transport http citehound https://getcitehound.com/api/mcp
Cursor
  1. Open .cursor/mcp.json in your project, or ~/.cursor/mcp.json for every project.
  2. Add this block.
  3. Restart Cursor, or reload its MCP connections.
.cursor/mcp.json 
{
  "mcpServers": {
    "citehound": {
      "url": "https://getcitehound.com/api/mcp"
    }
  }
}
Other clients

Any client that speaks Streamable HTTP can use the same address. By hand, send MCP-Protocol-Version: 2026-07-28 and Mcp-Method on every request, Mcp-Name on tools/call and prompts/get, and the protocol version and client capabilities in _meta; 2025-11-25 and 2025-06-18 are also accepted.

Limits

What to expect.

  • Read-only. Every tool reads. None writes to your site, sends a message or changes an account.
  • No signup. No key, no account, no login.
  • Nothing stored about you. The server keeps counters for an hour, keyed by the scanned domain. They hold no IP address, client name or account.
  • Live scans are limited. Three tools (scan_site, compare_sites, generate_schema) fetch the domain you name, at most six times per domain per hour. The other ten work from our own files.
  • Readiness, not presence. A score says whether AI crawlers can reach a site and whether its signals give a model reason to trust it. It does not say whether an assistant names you.
  • No live citation tool. A citation run is about 90 model calls, 18 questions asked five times each, and would exceed the function’s time limit. The question sets are free to try in your own assistant.

Read the privacy policy and the methodology.

Questions

The MCP server, briefly.

What is MCP?

MCP, the Model Context Protocol, is an open standard that lets an AI assistant call outside tools directly instead of a person copying results between a website and a chat. Connect Citehound’s server and the assistant can run a scan, build a fix or read a playbook inside the conversation.

Do I need an account?

No. There is no signup, no API key and no login. Add the server address to your client and ask.

Is it free?

All thirteen tools are free, and so are the three prompts. The three tools that scan a live site are limited per domain, so one site is not fetched over and over, and there is a ceiling across all callers.

What data does it fetch and keep?

The three tools that scan a live site (scan_site, compare_sites, generate_schema) fetch a domain’s public robots.txt, llms.txt, sitemap and homepage, and for a store one product page where robots.txt allows it. The other ten read our own files and fetch nothing. Nothing about who you are is stored. The server keeps counters keyed by a one-way hash of the scanned domain, which expire within an hour, to limit repeated fetches of one site.

Which clients work?

Any MCP client that supports servers over HTTP. Setup for Claude, Claude Code and Cursor is above. Custom connectors depend on your plan, so check your client’s help for the current steps.

Does it tell me whether AI names my brand?

No. The scan measures readiness: whether crawlers can reach your site and whether its signals give a model reason to trust it. It cannot observe what an assistant says, and a high score does not guarantee a mention. Two tools, get_citation_prompts and get_citation_sample, give you the questions and a real sample, so you can run the check yourself in your own assistant.

Prefer the browser?

Scan your site free