An MCP Website Service for Freelancers: Publish, Audit, Fix, Sign Off
Learn how Website Toolkit works as an MCP service to automate site audits, fixes, and launch checklists directly from your AI assistant.
Building a website has never been faster. An AI agent in Claude Code, Cursor or Windsurf can take a brief to a deployed staging site in an afternoon.
Everything after that is still slow. Has anyone checked the whole site? Are the broken links fixed? Is the client's feedback dealt with? Has everything on your launch list actually been done? That part still eats your evenings, and it's the part your reputation rests on.
Website Toolkit now works as an MCP website service. Your agent connects to it the same way it connects to any other tool, and once it has published a site it can carry on: audit the site, fix what the audit finds, work through your launch checklist and tell you what's left. One connection gives it the whole post-build workflow.
What MCP means here
MCP (Model Context Protocol) is the standard way AI assistants connect to outside services. Add Website Toolkit as an MCP server and your agent gets a set of website tools it can use whenever you ask in plain language. You never call a tool by name; you just ask: "Audit the site we just published and fix what you can."
Setup is one token and one command. Generate a token in your client portal under Domains → API access, then copy the command shown there:
claude mcp add --transport http wsc-crawler https://your-server/mcp \
--header "Authorization: Bearer your-token-here"
Other MCP clients take the same URL and header. If your tool doesn't support MCP, the same features are available as a plain REST API with the same token.
The token only covers the domains on your account. Your agent can't see anyone else's sites.
What your agent can do with it
1. Audit the site it just published
Once the site is live on staging or production, the agent can start a full audit itself. The crawler goes through every page and checks for:
- broken links, images, CSS and scripts
- missing titles, meta descriptions and headings
- slow pages
- spelling mistakes
- SEO warnings
The agent doesn't have to sit and wait. It checks back when the crawl finishes and pulls the results. It can also compare the live sitemap against the crawl, to catch pages you published but never linked to.
2. Fix the findings and close them off
The agent reads what's outstanding, with the page each issue was found on, then fixes the issues and marks each one done as it goes. Your audit report updates as it works, so the agent and the report always agree on what's left.
It can also do a quick spot check after it deploys a fix. It re-fetches every URL the audit flagged as broken and closes the ones that now load, without waiting for a full re-crawl.
"Fixed" doesn't stick on the agent's word. A fix mark applies to that audit only. Run the next audit and anything that wasn't really fixed comes straight back.
3. Work through your task lists
Most freelancers juggle more than one list for a site:
- The launch checklist: pre-launch and post-launch tasks, many of them ticked automatically from the crawl.
- Your to-do list: follow-ups, client requests, things you noticed.
- Client feedback: notes your client left on specific pages through the feedback widget.
Your agent can read all three. It can work through feedback items, moving each from not started to in progress to done, with a note on what changed. It can work through the launch tasks one by one. And when it spots something new, like a 404 in the footer or a form with no confirmation message, it adds a task to your list, flagged as AI-added, with the reason it gave.
4. Check every page on every device
The launch checklist's Site Content tab is a grid of every page on the site, with Desktop, Tablet and Mobile boxes before and after launch. On a large site, that grid is where checks quietly get missed.
Your agent can list the pages still unchecked, review each one at desktop, tablet and phone widths, and mark what it reviewed. It leaves a note on anything that looks wrong.
5. Re-check, and confirm nothing slipped
After the fixes are deployed, the agent starts a fresh audit and re-checks the launch checklist against it. Tasks the crawler can detect, like HTTPS, analytics, titles and broken links, update to match what the crawler actually found, not what the agent assumed.
For a site migration, the agent can also import every old URL, check that each one redirects to the right new page, and fix the ones that don't land.
6. The admin, too
The same connection also handles the paperwork around the work:
- logging the time the agent spent against your project
- putting the site on scheduled monitoring after launch
- telling you how much audit budget you have left
The part you keep
This is what makes it a service for freelancers rather than an autopilot: the agent does the work, and you do the sign-off.
When an agent marks a launch task done, the task doesn't tick. It becomes a proposal on your dashboard, with the agent's note explaining what it checked. You approve or dismiss it in one click. Pages the agent reviewed go into the same "awaiting approval" queue. A violet bar shows you everything waiting, and the Site Content filters show what's still unchecked, for example every page nobody has signed off on mobile yet.
There's a reason for this. An agent can confirm a page renders, but some things it can't confirm:
- that the contact form actually reaches your client's inbox
- that the prices on the site are right
- that the stock photos are licensed
- that the domain is registered in the client's name
Those are on you. Keep them as a Freelancer sign-off group in your launch template and they appear on every project, next to the agent's work, where they can't be forgotten.
Your client's review link shows only the checklist as you've approved it. They never see the agent's internals.
A day in practice
"We've just deployed clientsite.com. Run a full audit."
"Fix the broken links and missing meta descriptions, then close them off."
"Go through the client's open feedback, fix what you can, and note what you changed."
"Work through the pre-launch checklist. Propose each task done only after you've verified it, and add anything new you find to my to-do list."
"Review every page on mobile and tablet and mark what you checked."
"Re-audit, recheck the checklist, log the time, and tell me what's left for me."
Then you open the checklist, read the agent's notes, approve what's done, handle your own sign-off tasks, and send the client their review link.
Your agent built the site. With Website Toolkit connected, it also audits the site, fixes what the audit finds and works through your launch checklist, and you do the final sign-off.