Anonymous MCP server

A free SEO MCP server that needs no API key

10 tools over one stateless endpoint — read-only checks plus an account handoff: look up 63 AI crawlers from 22 providers, check what a robots.txt allows, inspect llms.txt and MCP discovery, read the Google update radar, and scan one page. No sign-up, no Bearer header — limits are counted per IP.

Works with Claude Code, Cursor, VS Code, Claude Desktop (via the npx bridge) and any Streamable HTTP client. No account data, no site changes, no crawling of your console.

Try asking your assistant

Crawler

“Which AI crawler is Mozilla/5.0 (compatible; GPTBot/1.2)?”

Access

“Can ClaudeBot and GPTBot read https://example.com/pricing?”

Readiness

“Is my site ready for AI agents? Score example.com.”

Each result carries its remaining budget and links for the docs, a shareable report and the account server — 5 URL-fetching calls a day is enough to answer a question, not to monitor a site.

Connect without a key

Point any Streamable HTTP client at the endpoint. There is no Authorization header, no OAuth dance and nothing to revoke — the server answers public data only.

Free MCP server URL

https://visdenda.com/v1/mcp/free
Claude Code: add the free server in one command
claude mcp add --transport http visdenda-free \
  "https://visdenda.com/v1/mcp/free"
Cursor or VS Code: URL only

Cursor (mcpServers):

{
  "mcpServers": {
    "visdenda-free": {
      "url": "https://visdenda.com/v1/mcp/free"
    }
  }
}

VS Code (servers, http type):

{
  "servers": {
    "visdenda-free": {
      "type": "http",
      "url": "https://visdenda.com/v1/mcp/free"
    }
  }
}
Claude Desktop: stdio bridge
npx -y @visdenda/mcp-free-bridge

Claude Desktop does not speak Streamable HTTP, so this zero-dependency bridge forwards stdio messages to the hosted endpoint. Cursor and VS Code can call the URL directly; per-client config lives in the npm package README.

{
  "mcpServers": {
    "visdenda-free": {
      "command": "npx",
      "args": [
        "-y",
        "@visdenda/mcp-free-bridge"
      ]
    }
  }
}
Verify from a terminal
curl -s https://visdenda.com/v1/mcp/free \
  -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

A successful call returns JSON with tools; malformed JSON returns -32700, an unknown tool -32602, and over-quota calls HTTP 429 with -32003 and Retry-After.

10 tools, one directory

Every tool returns a one-line summary, structuredContent and the remaining budget. Tool names are prefixed with visdenda_, so a result pasted into a chat still says where it came from. Schemas are served by tools/list — this page does not keep a second copy that can drift.

visdenda_search_ai_crawlers
No external fetch

Search 63 crawlers from 22 providers by name, vendor, category or robots.txt token.

visdenda_get_ai_crawler
No external fetch

Return one crawler's user agents, reverse-DNS suffixes, published IP ranges, sources and review date.

visdenda_match_user_agent
No external fetch

Paste a user-agent or a log line and get the matching crawlers, ranked by exact, prefix and substring hits.

visdenda_check_crawler_access
Fetches public URLs

Read a site's robots.txt once and report allow or blocked per crawler, the deciding rule and crawl-delay.

visdenda_check_llms_txt
Fetches public URLs

Check /llms.txt for a missing file, an HTML soft 404, a missing H1, broken links and the llmstxt.org shape.

visdenda_get_radar_updates
No external fetch

Recent P0/P1 Search updates with the number of active audit rules each one affects.

visdenda_check_mcp_server
Fetches public URLs

Probe the MCP server card, OAuth discovery documents, auth.md and the A2A agent card, with a fix per failure.

visdenda_check_agent_readiness
Fetches public URLs

Run up to eight agent-readiness probes on one site and return a 0–100 score with per-check detail.

visdenda_scan_page
Fetches public URLs

Run the public one-page scan (meta, canonical, robots, sitemap, llms.txt, schema, hreflang, redirects) and get a seven-day report link.

visdenda_connect_account
Starts a claim only

Create an agent registration claim: the user approves a six-digit code, the agent receives a one-hour token for the account server.

The directory covers 63 crawlers; per-entry sources and the license are on the crawler directory. Radar items come from the same public update radar.

Free checks vs your account data

Both endpoints speak Streamable HTTP. The difference is the data: the free server reads only what anyone can fetch from a URL, while the account server reads the properties you connected. On the account side a missing source reads na with a reason — never zero.

What the anonymous free MCP server and the account MCP server can read
CapabilityFree server · no keyAccount server · OAuth or key
Public checksCrawler lookup, robots.txt access, llms.txt, MCP discovery, agent readiness, one-page scan and the update radar.Site-scoped audit runs, findings and page reports. Public crawler lookups stay on the free server.
Search ConsoleNo. The free server has no access to a Search Console property.Clicks, impressions, CTR, top queries, landing pages and average position with prior-period deltas. Read-only OAuth; data lags 1–3 days.
GA4 and AI referralsNo. Sessions and conversions need a bound GA4 property or the account's page beacon.Sessions and engaged sessions per landing page, GA4 conversions and AI referral visits. GA4 page detail is a 7-day top-20 snapshot; a page outside it reads na.
Audits and findingsNo account history. A scan can preview one page, but findings are not saved to a site.Audit runs, rule-level findings, page reports and fix verification before and after a change.
Competitor changesNo.New, removed and changed competitor URLs from public sitemaps and page snapshots. No competitor traffic and no competitor AI citations.
Write toolsRead-only. Nothing can be started or changed.run_audit and import_sitemap need an API key with audits:write, confirm:true and an idempotency key. OAuth connections stay read-only.
History and limitsPer-IP daily caps; a scan report link lasts seven days and is not attached to an account.Runs and findings persist. MCP reads are free on both plans with a per-account daily cap and do not consume the REST read quota.

The account side is the per-page join: page decisions and reporting, Search Console performance, traffic and conversions and competitor changes. The free server stays a public check, not a trial of that data.

Page-decision prompts for the account server

These need Search Console and GA4 joined to a page. The free server cannot answer them, because it cannot see either property.

  • “Which page should I fix first?”

  • “Why do clicks not match sessions on /pricing?”

  • “Did the fix work?”

Connect a property with the Search Console guide. Setup for the account endpoint is in the MCP guide.

What the free tier allows

Limits are counted per client IP and reset at 00:00 UTC. There is no account to create and nothing to rotate; a shared assistant egress (for example claude.ai) draws on a separate pool with its own daily ceiling so one client cannot exhaust the service.

100per day

Directory and radar calls

Crawler search, single-record reads, user-agent matching and update radar over 63 entries. No external request.

5per day

Public URL checks

robots.txt access, llms.txt, MCP discovery and agent readiness. A probe may fetch one to three URLs; robots.txt and llms.txt responses are cached for an hour at the edge.

3per day

Single-page scans

The same quota as the public free scan tool, and one of the URL-check calls above. The report is shareable for seven days, and opening it signed in lets you save it to a site.

30per minute

Short-term burst

A minute-level bucket answers bursts before the daily budget is touched; malformed calls count too.

Every result carries the budget

Successful and failed results include _meta["dev.visdenda/free"] with the remaining calls, the cost class, whether a retry can help, and links to the docs, the report and the account server. A model can decide when to stop without guessing.

What a 429 looks like

HTTP 429 with a JSON-RPC -32003 error, data.code of MCP_DAILY_LIMIT or RATE_LIMITED, and Retry-After in seconds. Quota counts are keyed by a hash of the IP; scan reports keep the scan result for seven days under the shareable link.

Need more than a handful of URL checks a day? The account server has no per-IP budget and reads data the free tier cannot see.

When one check becomes a habit

The free server answers one question about public data. The account server answers the same questions about your own property — and remembers the answers.

Step 1

Ask the agent to connect

Call visdenda_connect_account with the account owner's email. It returns a claim URL, a six-digit code and a claim token; nothing is charged and no site is touched.

Step 2

Approve the code

Open the claim URL, sign in and approve the scopes the task needs — radar:read, rules:read, audit:read, sites:write. The code expires in ten minutes; the registration itself in 24 hours.

Step 3

Keep working with a token

The agent polls the token endpoint with the claim token and receives a one-hour access token for the account server. Access is revocable under Access → Connected apps.

Search Console and GA4

Clicks, impressions, CTR, position and sessions per page, joined to the same site and page as the audit.

Saved history

Audit runs, findings and index status persist. The free scan disappears after seven days; an account keeps the record.

Monitoring and radar alerts

Scheduled checks and per-site update impact, instead of one-off answers.

Reads stay free

MCP read tools are free on both plans with a per-account daily limit; they never consume the REST read quota. Pro adds verified sites and audit capacity.

Claim links are also embedded in every free result, so a model can offer the handoff without being told to.

Questions the free server gets

Do I need an API key or an account?

No. The free server takes no Authorization header, so there is nothing to create, copy or rotate. Any client that speaks Streamable HTTP can connect to the endpoint directly.

What are the limits?

Calls are counted per client IP and reset at 00:00 UTC: 100 directory and radar calls, 5 calls that fetch public URLs, and 3 page scans per day. A short-term bucket also caps 30 requests a minute. Over a limit the server returns HTTP 429 with a JSON-RPC -32003 error and Retry-After, plus the remaining budget in every successful result.

Does it read my Search Console or GA4 data?

No. The free server only reads public data — the crawler directory, the public update radar and URLs you pass in — and it never writes to a site. Search Console and GA4 reads, saved history and the two opted-in write actions live in the account server.

Can I reuse the crawler data?

Yes. The crawler directory is published under CC BY 4.0; attribute Visdenda and link to /crawlers. Every crawler result also carries its official sources, so you can cite the vendor page instead of the tool.

How do I use it in Claude Desktop?

Claude Desktop speaks stdio, so run the zero-dependency bridge with npx: it forwards messages to the same hosted endpoint. Cursor, VS Code, Claude Code and other Streamable HTTP clients can use the endpoint URL directly without the bridge.

What happens to a free scan report?

The scan returns a shareable report URL that stays available for seven days. It is not indexed and is not attached to any account. Opening the report while signed in lets you save the findings to a site, which keeps them after the preview expires.

Why is there a free server at all?

Because most agent questions stop after one check: which bot is this, can it read the page, does the site expose llms.txt. Charging for that answer would hide the data. When the question becomes continuous — history, Search Console, alerts — the account server is the next step, and the free results link to it.

Machine-readable: llms.txt, ai-catalog.json and crawlers.json. Account server: setup guide.