dataforseo-mcp-skill

Languages: English · Français

A Claude skill (Claude Code / Claude Desktop) that acts as a cost-aware playbook for the DataForSEO API, designed to work in symbiosis with the official DataForSEO MCP server.

The MCP gives Claude the hands (tooling access to the whole API). This skill gives it the brain: which endpoint for which need, which execution mode, and above all how to not burn budget (DataForSEO is a prepaid API, billed per request).

The problem it solves

The official MCP exposes 4 clever tools (docs_index, docs_list_sections, docs_search, api_request) that cover the entire v3 API. Powerful — but Claude then faces hundreds of endpoints, several execution modes at very different prices, and no cost guardrails. The result: either it under-uses the API, or it fires off expensive batches without warning.

This skill is the missing knowledge layer:

┌──────────────────────────────┐     ┌───────────────────────────────┐
│  DataForSEO MCP (the hands)  │     │  This skill (the brain)        │
│  • docs_*  → browse the docs │  +  │  • which endpoint for which job │
│  • api_request → call anything│     │  • Live vs Standard vs Labs     │
│                              │     │  • COST guardrails + sandbox    │
└──────────────────────────────┘     │  • ready-to-use recipes         │
                                      └───────────────────────────────┘

Advantages

• 💸 Cost-aware first. DataForSEO is prepaid and billed per call. The skill enforces checking the balance (appendix/user_data, free), warning + estimating before any large batch, testing request structure in the free Sandbox before production, and reporting the cost after every real call.
• 🧭 The right endpoint, the right mode. A decision map across SERP, Keywords Data, Labs, Backlinks, On-Page, Content/Domain Analytics, Merchant/App/Business — and crucially when to prefer DataForSEO Labs (internal database, ~10× cheaper) over live SERP scraping for keyword/competitor research.
• ⚙️ Live vs Standard queue vs Labs. The skill explains the latency/cost trade-off (.../live expensive and immediate, task_post+task_get ~2× cheaper but async, Labs cheap with no scraping).
• 📋 Ready-made recipes. Search volume, SERP rank tracking, keyword ideas, competitors / domain intersection, bulk backlinks, on-page audit — as MCP calls or as curl (reusable in n8n, Make, scripts).
• 🔌 No rigid coupling. Unlike a "frozen Python wrappers" skill, it hardcodes no subset of endpoints: it leans on the MCP's api_request, so the whole API stays reachable.

Prerequisite: the official MCP

This skill assumes the DataForSEO MCP is wired into your Claude client.

Claude Desktop — in claude_desktop_config.json:

{
  "mcpServers": {
    "dataforseo": {
      "command": "npx",
      "args": ["-y", "dataforseo-mcp-server", "--mode", "stdio"],
      "env": {
        "DATAFORSEO_LOGIN": "<your_api_login>",
        "DATAFORSEO_PASSWORD": "<your_api_password>"
      }
    }
  }
}

Claude Code — claude mcp add dataforseo -- npx -y dataforseo-mcp-server --mode stdio (then set the environment variables).

login = your API account email, password = the API password from https://app.dataforseo.com/api-access (plain values — the MCP builds the Basic Auth header itself).

The skill also works without the MCP via a curl fallback (see skill/references/recipes.md), but the MCP + skill combo is the recommended usage.

Installing the skill

Copy the skill/ folder into your client's skills directory, named dataforseo/:

# Claude Code / Desktop (user level)
cp -r skill "$HOME/.claude/skills/dataforseo"

Claude triggers the skill whenever a task touches SEO / SERP / keywords / backlinks / on-page audits, or whenever you want to test a DataForSEO request.

Compatibility beyond Claude

Three layers, from most Claude-specific to fully portable:

Layer	Compatibility
Skill format (SKILL.md + auto-triggering frontmatter)	Claude ecosystem (Claude Code, Claude Desktop, Agent SDK). The auto-trigger mechanism is Claude-specific.
DataForSEO MCP server	Any MCP client — Cursor, Cline, Windsurf, Zed, VS Code, etc. MCP is an open standard, no longer Claude-only.
The content (endpoints, cost/modes, curl recipes)	Universal. Plain Markdown + curl calls: reusable as a rules/context file for any agent (e.g. .cursorrules), or directly in n8n / Make / scripts.

In short: the packaging is optimized for Claude, but the substance is portable. A Cursor user can wire up the same MCP and paste the playbook content as rules. PRs adapting it to other agents are welcome.

Structure

skill/
├── SKILL.md                    # main playbook: endpoint + mode decision, cost rule
└── references/
    ├── cost-and-modes.md       # Live vs Standard queue vs Labs, free sandbox, trade-offs
    ├── endpoints.md            # full v3 endpoint map (exact paths)
    └── recipes.md              # ready-made recipes (MCP + curl), incl. common SEO patterns

The playbook is written in French

The skill content is currently written in French. It is easily translatable — PRs for an English version are welcome.

License

MIT — see LICENSE.

Credits

• API & MCP: DataForSEO · mcp-server-typescript
• Inspiration (standalone skill approach): nikhilbhansali/dataforseo-skill-claude
• Maintained by Creapulse.