MCP Plans: $30 a Month for Web Data Access — Not a Browser
Update. The single flat plan this post originally announced has been retired. Anyone who subscribed to it keeps it at the same price; new signups pick from the MCP plans described below. The rest of the post — how the server works, and why it isn't a browser — still holds.
If you've been using Claude Desktop, Claude Code, or Cursor with MCP servers, you know the problem: you're constantly doing credit math. Should you look up one more company? Can you afford to pull that Reddit thread? Is this search worth another 10 credits?
The MCP plans are flat monthly: MCP at $30, MCP x5 at $99, MCP x15 at $199. Every tier reaches the whole catalogue — tiers differ in how much work they cover, never in what they can see. There is no per-request meter to watch mid-conversation.
This is not browser automation. This is not Puppeteer or Playwright. You are not parsing HTML. You ask in plain English and JSON comes back — business data, Instagram posts, Reddit threads, Twitter/X feeds, or any URL — delivered into your AI client.
What you get
The server exposes five meta-tools. Not one per endpoint — five:
discover— find the right source and endpoint for what you're asking about, across the whole catalogue.execute— call that endpoint with parameters and get JSON back.get_page— read any URL that no dedicated source covers.query_cache— filter, sort and aggregate what you already pulled, without fetching it again.export_data— hand a result set off as a file instead of pushing every row through the conversation.
An earlier version of this server registered a separate tool for every endpoint. That design does not survive a catalogue this size: dozens of tool definitions eat your context window before you have asked anything, and they go stale every time the catalogue grows. Five meta-tools stay constant. New sources simply show up through discover — you don't reconnect, and nothing in your setup changes.
You connect once to mcp.anysite.io. After that you ask in plain English and the model picks the endpoint.
How it works (not a browser)
MCP (Model Context Protocol) is how an AI client reaches an outside system. Instead of writing extraction scripts or driving a headless browser, you ask for what you need — for example, to find machine-learning engineers at Google with transformer experience.
Claude calls discover to locate the right business-data endpoint, calls execute, and gets records back as JSON:
{
"profile": {
"name": "Jane Smith",
"headline": "ML Engineer @ Google",
"skills": ["Transformers", "PyTorch", "NLP"],
"experience": [...]
}
}No page loads in your environment. No DOM. The work happens server-side, and what arrives is a JSON record you can pass straight to the next step.
What this is not
This is not a browser extension. This is not Selenium or headless Chrome. You are not automating clicks, filling forms, or simulating a user session.
Think of it as web APIs reached through MCP rather than browser automation: data that is ready to use, not markup that needs parsing.
What an MCP plan does not cover
An MCP plan reaches one thing: the Remote MCP Server, from an MCP client. It does not open the REST API, and it does not open the CLI. If your application code calls api.anysite.io directly, or you run jobs from a terminal, you need a plan on the API line — those start at $49 a month and bill per credit.
Access runs one way. An API key also works inside MCP clients, billed per credit like the rest of that plan, so from $49 the same key covers REST, the CLI and MCP together. A key from an MCP plan does not work in the other direction.
Getting connected
Sign up at anysite.io and open your dashboard. There are one-click install buttons for Claude and Cursor. For other MCP clients, copy the connection URL from the same screen.
Who this is for
You're building agents that need live web data. You're doing competitive research from Claude Desktop. You're prototyping in Cursor against real business data instead of fixtures.
The case for a flat plan is simple: research is bursty. You spend a quiet week, then one afternoon you want forty lookups in a row while a thought is still fresh. A meter turns that afternoon into arithmetic. A flat plan doesn't, so you ask the question instead of pricing it.
Two skills to start with
Two pre-built skills work on every MCP plan.
Person analyzer. Give Claude a name and a company. It pulls the business profile, Twitter/X activity, Reddit posts and general web presence, then returns a read on professional background, public opinions, technical depth and communication style. Useful for recruiting, partnership checks or competitive work.
Competitor analyzer. Give Claude a company name or domain. It pulls the company profile, employee profiles, social presence, site content and public discussion, then returns organizational shape, hiring trends, product positioning and market strategy. Useful for sales research, market analysis or diligence.
Both chain a sequence of discover and execute calls behind a single command, so you don't orchestrate ten queries by hand. Everything comes back as JSON, ready to analyze.
Throughput and fair use
There is no per-request counter on an MCP plan and nothing to top up mid-month. Throughput is sized to the tier and governed by fair use, which ordinary research and agent work does not run into.
If you do reach the capacity of your plan, nothing breaks: your agent is told so in the conversation and can pick up again later. If that starts happening regularly, that's the signal to move up a tier, and upgrades apply immediately. For a specific workload, write to hello@anysite.io.
How to get started
- Sign up at anysite.io.
- Pick an MCP plan — MCP, MCP x5 or MCP x15. Each starts with a 7-day free trial with full plan access.
- Connect your MCP client (Claude Desktop, Claude Code, Cursor, Windsurf, Cline).
- Ask for what you need in plain English.
Full setup instructions: https://docs.anysite.io
If you're on a credit plan today, you can switch lines from your account dashboard, and switch back.
Technical details
The server runs at mcp.anysite.io for OAuth connections and mcp.anysite.io/mcp for key-based ones. Both speak the Model Context Protocol, and the five meta-tools register themselves when you connect.
OAuth covers clients like Claude Desktop and Cursor; clients that expect a key use the direct endpoint. Keep connection URLs and keys out of version control.
Responses follow MCP conventions: JSON out, input validation, and readable errors that say what went wrong. Extraction happens server-side, so nothing renders in your environment.
What's next
Any URL already works: get_page handles pages that no dedicated source covers, so you're not limited to the platforms with first-class endpoints. New sources land in the catalogue continuously and appear through discover on their own — no reconnect, no config change.
Questions
Can I use this with more than one AI client?
Yes. The same MCP connection works in Claude Desktop, Claude Code and Cursor at once.
What if I need the REST API or the CLI too?
That's the API line, from $49 a month, billed per credit. One API key covers REST, the CLI and MCP clients together.
Do MCP plans use credits?
No. Flat monthly, nothing to count and nothing to top up.
Can I share an account?
One account per person. For team access, write to hello@anysite.io.
Can I cancel anytime?
Yes, from your dashboard. Access runs to the end of the billing period.
Is this browser automation?
No. JSON responses, not HTML pages. No browser involved.
Summary
MCP plans are flat monthly — $30, $99 or $199 — and they reach the whole catalogue through five meta-tools from any MCP client. No credit counting. The REST API and the CLI are a separate line, from $49 a month.
What you get back is JSON, not a rendered page, and it lands in the conversation you're already having.
Start for free at anysite.io, or read the documentation at docs.anysite.io.