Disclosure: I built and operate the tools in this post (SiteDoctor, VerifAI and RugShield, run by ManHourPlan). They have a free tier and paid credit packs. Everything below was run against the live server while writing this.
Coding agents are good at shipping pages and bad at checking them. They'll write a landing page, call it done, and never notice the missing meta description, the 77-character title or the absent security headers.
The fix is to give the agent a tool it can call to check its own work. With a remote MCP server, that takes one config entry: no package to install, no Docker, no API dashboard.
What "remote MCP" means
The Model Context Protocol is the standard way AI clients (Claude, Cursor, VS Code Copilot and others) discover and call external tools. An MCP server can run locally as a process, or remotely over HTTP ("Streamable HTTP" transport).
A remote server is just a URL. Your client connects to it, asks tools/list, and the model can then call those tools like functions. Nothing runs on your machine except the client you already have.
The server in this post is SiteDoctor:
https://manhourplan.com/doctor/mcp It's also listed in the official MCP Registry as com.manhourplan/sitedoctor.
Step 1: Add it to your client
Claude Code
claude mcp add --transport http sitedoctor https://manhourplan.com/doctor/mcp Cursor (~/.cursor/mcp.json, or .cursor/mcp.json in a project)
{ "mcpServers": { "sitedoctor": { "url": "https://manhourplan.com/doctor/mcp" } } } VS Code (.vscode/mcp.json)
{ "servers": { "sitedoctor": { "type": "http", "url": "https://manhourplan.com/doctor/mcp" } } } Or from a terminal:
code --add-mcp '{"name":"sitedoctor","type":"http","url":"https://manhourplan.com/doctor/mcp"}' Claude Desktop (claude_desktop_config.json)
Claude Desktop's config file launches local processes, so this uses the mcp-remote bridge (needs Node.js):
{ "mcpServers": { "sitedoctor": { "command": "npx", "args": ["-y", "mcp-remote", "https://manhourplan.com/doctor/mcp"] } } } If your plan supports custom connectors, you can instead paste the URL under Settings → Connectors.
Restart or reload the client and you should see these tools:
| Tool | Cost | What it does |
|---|---|---|
register_free_key | free | Creates a key with 25 credits. No email, no payment. |
scan_website | 3 credits | Full audit: SEO, performance, security headers, SSL, mobile, content, broken links. 0–100 score and grade. |
check_page_seo | 1 credit | Fast on-page SEO: title, meta description, headings, alt text, canonical, Open Graph, robots.txt, sitemap.xml |
check_broken_links | 2 credits | Crawls a page and tests up to 50 links |
check_credits | free | Remaining balance |
Step 2: Get a free key
Ask your agent:
Register a free SiteDoctor key and remember it for this project.
The agent calls register_free_key. Real response (key redacted, top-up text trimmed):
{ "api_key": "<redacted>", "credits": 25, "message": "Free key created. Credits: scan_website=3, check_broken_links=2, check_page_seo=1." } 25 credits is about 8 full scans, or 25 quick SEO checks. Every scanning tool takes the key as an api_key argument, so the agent needs to keep it. Put it in your project notes, a rules file, or just in the prompt.
Step 3: Let the agent audit a site
Run scan_website on https://manhourplan.com and tell me what to fix first.
Here's the real output from scanning our own homepage while writing this (trimmed slightly):
{ "url": "https://manhourplan.com", "score": 91, "grade": "A", "categories": { "seo": 28, "performance": 22, "security": 21, "mobile": 10, "content_links": 10 }, "max_per_category": { "seo": 30, "performance": 25, "security": 25, "mobile": 10, "content_links": 10 }, "metrics": { "response_ms": 153, "html_kb": 26.6, "word_count": 1289, "images_missing_alt": 0, "https_valid": true, "ssl_days_remaining": 58 }, "links": { "checked": 12, "broken": [] }, "issues": [ { "cat": "security", "sev": "medium", "msg": "Missing header: CSP" }, { "cat": "seo", "sev": "low", "msg": "Title length 77 chars (ideal 30-60)" }, { "cat": "performance", "sev": "low", "msg": "No Cache-Control header on HTML" } ] } So yes, the tool's author's own site has three things to fix. That output is the useful part. Because it's structured JSON, the agent doesn't just report it, it can act on it:
- Shorten the
<title>to under 60 characters. - Add a
Content-Security-Policyheader in the web server config. - Set
Cache-Controlfor HTML responses.
Then call scan_website again to confirm the fixes. That loop (build → scan → fix → rescan) is the main reason to give an agent this tool instead of running a one-off audit yourself.
A pattern that works well
Add a line like this to your agent rules (CLAUDE.md, .cursorrules, Copilot instructions):
After changing any public page, run check_page_seo on its URL (SiteDoctor MCP, api_key in .sitedoctor-key). Fix medium/high issues before saying the task is done. check_page_seo costs 1 credit and skips the link crawl, so it's cheap enough to run after every deploy. Save scan_website for a fuller pass before launch.
Step 4 (optional): Try it with curl
If you want to see the protocol itself, it's plain JSON-RPC over HTTP:
U=https://manhourplan.com/doctor/mcp H='-H content-type:application/json -H accept:application/json,text/event-stream' # open a session SID=$(curl -s -D - -o /dev/null $H -X POST $U -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-03-26","capabilities":{},"clientInfo":{"name":"curl","version":"1"}}}' \ | grep -i '^mcp-session-id' | cut -d' ' -f2 | tr -d '\r') curl -s $H -H "mcp-session-id: $SID" -X POST $U -d '{"jsonrpc":"2.0","method":"notifications/initialized"}' >/dev/null # mint a free key curl -s $H -H "mcp-session-id: $SID" -X POST $U \ -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"register_free_key","arguments":{}}}' # quick SEO check (1 credit) curl -s $H -H "mcp-session-id: $SID" -X POST $U \ -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"check_page_seo","arguments":{"url":"https://example.com","api_key":"YOUR_KEY"}}}' Two sibling servers
The same setup (remote URL, register_free_key, 25 free credits) works for two other servers I run:
- VerifAI: email verification (syntax, disposable-provider and MX checks) plus domain intel (DNS, RDAP registration, SSL days left) for lead lists and signup forms.
https://manhourplan.com/mcp· details - RugShield: Solana token risk scans (0–100 risk score, mint/freeze authority, holder concentration, liquidity, pair age).
https://manhourplan.com/shield/mcp· details
Each server has its own key and credit balance.
Pricing
- Free: 25 credits per server, no signup.
- Paid credit packs (card via Stripe, credited automatically; crypto also accepted):
Those links are for SiteDoctor. Each product page has its own. You can pass your existing api_key at checkout to top it up, or leave it blank and a new key is issued. Billing questions: support@manhourplan.com.
Limits worth knowing
- It audits one URL per call and samples up to 50 links. It isn't a full-site crawler like Screaming Frog.
- The performance score comes from server response and page weight. It isn't a Lighthouse run in a real browser.
- The scanner fetches public URLs only, so it can't audit localhost or pages behind a login.
For "did my agent just ship something broken?" checks, that has been the right trade-off. For a deep audit of a large site, use a dedicated crawler.
Full reference, including all four tools from this studio: manhourplan.com/tools. If something in the output looks wrong, I'd like to hear about it.
