XiaoFlow MCP Server (xiaoflow-mcp-server)

Official Model Context Protocol (MCP) server for XiaoFlow AI SEO and keyword intelligence tools.

Connect Large Language Models (LLMs) such as Claude Desktop, Cursor, Windsurf, and VS Code directly to XiaoFlow's search engine optimization, keyword discovery, and domain analytics.

Remote endpoint: https://mcp.xiaoflow.com/mcp

Transport: MCP Streamable HTTP with OAuth 2.1 / PKCE, plus stdio through the npm package

Safety: Read-only research tools, except asynchronous expansion task creation; no destructive tools

Official listings

• npm — xiaoflow-mcp-server

• Smithery — xiaoflow/xiaoflow-mcp

• Glama — xiaoq-in/xiaoflow-mcp

• GitHub — xiaoq-in/xiaoflow-mcp

• Full setup guide and prompt templates

Related MCP server: keyword-steroids-70

✨ Features & Capabilities

• 🔍 Keyword Discovery: Generate high-intent search keywords and SEO ideas from keyword, URL, or domain seeds.

• 📊 Domain Analytics: Analyze domain-level search performance, organic traffic metrics, and keyword distributions.

• 📈 Search Trend Analysis: Compare keyword demand, competition, CPC, and historical trends.

• 🔒 Flexible Authentication: Supports API Key authentication via query parameters (?key=), Bearer tokens, environment variables, or Web OAuth login.

⚡ Quick Start

1. Run via npx (stdio)

Run the server directly using npx:

npx -y xiaoflow-mcp-server

Pass your XiaoFlow API key via environment variable:

XIAOFLOW_API_KEY="YOUR_API_KEY" npx -y xiaoflow-mcp-server

2. Connect via Streamable HTTP with web login (recommended)

Use the canonical remote endpoint in clients that support remote MCP. The client discovers XiaoFlow OAuth automatically and opens the browser for login and consent:

https://mcp.xiaoflow.com/mcp

Legacy clients can still use https://mcp.xiaoflow.com/sse?key=YOUR_API_KEY.

💻 Client Integration Guides

Cursor Setup

Add XiaoFlow MCP to Cursor:

• Name: xiaoflow

• Type: http

• URL: https://mcp.xiaoflow.com/mcp

Or click Add to Cursor directly on the XiaoFlow MCP Portal.

Claude Desktop Setup

Add the following entry to your claude_desktop_config.json:

{
  "mcpServers": {
    "xiaoflow": {
      "command": "npx",
      "args": ["-y", "xiaoflow-mcp-server"],
      "env": {
        "XIAOFLOW_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

Windsurf, VS Code, and other remote clients

Use native HTTP configuration where available:

{
  "mcpServers": {
    "xiaoflow": {
      "type": "http",
      "url": "https://mcp.xiaoflow.com/mcp"
    }
  }
}

For stdio-only clients, bridge to the OAuth-enabled remote endpoint:

{
  "mcpServers": {
    "xiaoflow": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.xiaoflow.com/mcp"]
    }
  }
}

🛠️ Available MCP Tools

Every tool publishes:

• descriptions for every input parameter;

• a named JSON output schema, including success and error fields;

• MCP safety annotations (readOnlyHint, destructiveHint, idempotentHint, and openWorldHint);

• a human-readable tool title for MCP clients and directories.

Example prompts

Get US English metrics and 24 months of history for "AI SEO tools".

Find every related keyword for "standing desk", 200 per page, and continue
until has_more is false. Return search volume, CPC, competition, intent, and history.

Compare these 1,000 keywords over 48 months and rank them by search volume growth.

Expand "home office" for four rounds, keep terms with at least 100 monthly
searches, and poll the task until it is complete.

🔑 Authentication

Obtain your API key from the XiaoFlow MCP Dashboard.

Supported authentication methods:

1. Web Login OAuth (recommended): connect to https://mcp.xiaoflow.com/mcp; compatible clients discover OAuth, PKCE, and dynamic client registration automatically.

2. Environment Variable: set XIAOFLOW_API_KEY when running via npx.

3. Authorization Header: send Authorization: Bearer YOUR_API_KEY.

4. Legacy Query Parameter: append ?key=YOUR_API_KEY to the legacy SSE URL.

🐳 Docker / Glama

The repository includes a production multi-stage Dockerfile for directory build verification and stdio deployment:

docker build -t xiaoflow-mcp .
docker run --rm -i \
  -e XIAOFLOW_API_KEY="YOUR_API_KEY" \
  xiaoflow-mcp

The image runs as the unprivileged Node user, excludes local secrets and build state, and writes MCP protocol messages only to stdout.

🔐 Security and data handling

• OAuth login occurs only on www.xiaoflow.com; MCP clients never receive your password.

• API keys and OAuth tokens are sent only to the configured XiaoFlow API endpoint.

• The server does not scan or upload project files.

• Tool calls query external XiaoFlow/Google Ads-backed data and may consume account credits.

• No tool deletes or modifies keyword or domain data.

Report vulnerabilities privately through the repository owner or the XiaoFlow contact page. Do not include tokens or customer data in public issues.

✅ Quality and compatibility

• MCP protocol: Streamable HTTP and stdio

• Authentication: OAuth 2.1 with PKCE, Bearer API key

• Tool schemas: parameter descriptions, structured output schemas, annotations

• Optional discovery methods: resources and prompts return valid empty lists

• Runtime: Node.js 18+ (node:20-alpine in Docker)

• Continuous endpoint: https://mcp.xiaoflow.com/mcp

📄 License

MIT © XiaoFlow

Complete bulk results and files

bulk_keyword_metrics return all queried rows (up to 1,000 per request), rather than a 12-row report preview. Hosted MCP responses include CSV and JSON resource links plus structuredContent.exports. Download links are unguessable, valid for 24 hours, and return an attachment without requiring a separate API query. Historical trends are disabled by default; set with_history: true to request them. CSV contains the requested keyword metrics; JSON preserves the entire original result. Save the download itself to keep every row, regardless of the chat client's display limits.

The stdio server returns the full structured result and embedded text/csv and application/json resources, which clients can save locally. A remote MCP server cannot choose a path on your computer; the client saves the downloaded/resource file at your chosen location.

Regression checks: npm run build && node --test tests/keyword-export.test.mjs. To exercise hosted export storage and downloads locally, run Wrangler on port 8799 with XIAOFLOW_API_URL=http://127.0.0.1:8899, then node tests/worker-export.integration.mjs.

MCP results are now raw JSON, without chart images or dashboard widgets. Bulk metrics have no MCP total-keyword ceiling: requests are split into API batches, and discovery/domain keyword listings follow every reported result page. Exports preserve all rows, monthly histories and additional metric fields. The previous 8 MB export cap is removed. Upstream data availability, account credits and platform transport/storage limits still apply; these are not silently bypassed.