Model Context Protocol · Free tier

The web intelligence
MCP server.

Give your AI agent the whole ToolTrace toolkit in one install. Scrape any page to clean Markdown, read metadata and structured data, run SEO audits, detect what a site is built with, and check XML sitemaps. Works with Claude, Cursor, VS Code, Codex and any other MCP client.

Quick start

Two steps.

Get a free key at tooltrace.io/signup, then install the server. The free plan includes 1,000 credits a month.

Install · npm
npm install -g @tooltrace/mcp-server

MIT licensed. Read the source on GitHub or the package on npm.

The toolkit

What your agent can do.

Every tool returns structured JSON with fetch evidence attached, so an agent can tell you not just what it found but which page it actually read.

tooltrace_extract

Turn a webpage into clean Markdown, plain text, structured sections, metadata, links and JSON-LD in one call.

tooltrace_metadata

Read the title, description, canonical URL, language, author, publication date, Open Graph and Twitter Card fields.

tooltrace_links

Extract internal and external links with anchor text, rel attributes and nofollow classification.

tooltrace_schema

Pull JSON-LD structured data from a page, normalized and deduplicated.

tooltrace_seo_audit

Run twelve on-page SEO checks with severity, evidence and a specific recommendation for each.

tooltrace_tech_stack

Detect the CMS, frameworks, analytics, CDN, hosting and other technologies behind a site, with evidence.

tooltrace_sitemap

Validate an XML sitemap and check the URLs inside it for status, redirects, robots rules and indexability.

Configuration

Connect your client.

Every MCP client runs the same command with the same environment variable. If yours is not listed, use tooltrace-mcp as the command and set TOOLTRACE_API_KEY.

Claude Code
claude mcp add tooltrace tooltrace-mcp -e TOOLTRACE_API_KEY=your-api-key
Claude Desktop · claude_desktop_config.json
{
  "mcpServers": {
    "tooltrace": {
      "command": "tooltrace-mcp",
      "env": { "TOOLTRACE_API_KEY": "your-api-key" }
    }
  }
}
Cursor · .cursor/mcp.json
{
  "mcpServers": {
    "tooltrace": {
      "command": "tooltrace-mcp",
      "env": { "TOOLTRACE_API_KEY": "your-api-key" }
    }
  }
}
VS Code · .vscode/mcp.json
{
  "servers": {
    "tooltrace": {
      "command": "tooltrace-mcp",
      "env": { "TOOLTRACE_API_KEY": "your-api-key" }
    }
  }
}
Codex · ~/.codex/config.toml
[mcp_servers.tooltrace]
command = "tooltrace-mcp"
env = { TOOLTRACE_API_KEY = "your-api-key" }
Windsurf · mcp_config.json
{
  "mcpServers": {
    "tooltrace": {
      "command": "tooltrace-mcp",
      "env": { "TOOLTRACE_API_KEY": "your-api-key" }
    }
  }
}

In practice

Things you can ask.

Once connected, the tools are available in normal conversation. Your agent decides which to call.

Why web scraping through MCP

An agent asked to research a page will otherwise guess from its training data or fetch raw HTML and burn tokens parsing it. A web scraping MCP server gives it clean, structured input instead, and fetch evidence it can cite.

That evidence matters more for agents than for people. A response that records the requested URL, the final URL after redirects, the HTTP status and the render method lets an agent tell you when it read a different page than the one you named.

Built for retrieval

Markdown output is chunk-ready, links resolve to absolute URLs, and every response carries a content hash for change detection.

An SEO MCP server too

Several of the tools are technical SEO. An on-page SEO audit with twelve weighted checks, tech stack detection, and XML sitemap checking with per-URL indexability. Ask an agent to audit a page or a sitemap and it has the tools to answer with evidence rather than opinion.

Compliant by default

Access challenges are detected and reported, never defeated. No CAPTCHA solving, no bypassing access controls, so it is safe inside a commercial product.

Questions

MCP server questions.

What is an MCP server?

Model Context Protocol is an open standard for connecting AI assistants to external tools and data. An MCP server exposes a set of tools that a client such as Claude, Cursor, VS Code or Codex can call during a conversation. The ToolTrace server exposes one tool per ToolTrace endpoint.

Do I need an API key?

Yes. Sign up at tooltrace.io for a free key. The free plan includes 1,000 credits per month, which is enough for regular use across every tool.

Which MCP clients are supported?

Any client that speaks the Model Context Protocol over stdio. Claude Code, Claude Desktop, Cursor, VS Code, Codex and Windsurf are covered below, and every other client takes the same command and environment variable.

Can I use it as an SEO MCP server?

Yes. Several of the tools are technical SEO: an on-page audit with twelve weighted checks returning severity and evidence, tech stack detection, and XML sitemap checking that verifies status, redirects, robots rules and indexability for each listed URL. An agent can run a full page or sitemap audit and cite what it found.

Does it render JavaScript pages?

Yes. Extraction, metadata, links, schema, SEO audit and tech stack detection all start with a fast static fetch and use a real browser only when the page requires it.

How much does each tool call cost?

A static call costs 1 credit and a browser-rendered call costs 5. Sitemap checks cost 1 credit per 10 URLs inspected. Failed calls cost nothing.

Is it open source?

The server is MIT licensed and published on npm as @tooltrace/mcp-server. The ToolTrace API behind it is a hosted service.

Also available

Other ways to connect.

Beyond MCP, ToolTrace ships official packages for Python, LangChain and LlamaIndex, plus a REST API and marketplace listings. All of them are open source under the ToolTrace-io organisation. See the documentation for setup guides.