We built an MCP keyword-intelligence server — real-time SEO data, zero dashboard overhead

Any SEO consultant knows the ritual: open the tool, create a project, type a keyword, wait, export, then repeat across three more tools and reconcile everything in a spreadsheet. The data is good. The access pattern is the bottleneck.

So we built a faster path — an MCP server that lets Claude Code query keyword metrics directly, in natural language, and hand back structured results in seconds. This post documents the loop end to end, with real output from a live session.

01Why manual keyword research breaks down

Volume lives in one tool, SERP data in another, trends somewhere else again. A single keyword group can take 45–90 minutes before a recommendation is written — most of it spent moving numbers between tabs rather than making decisions.

02What an MCP server changes

MCP stands for Model Context Protocol — an open standard that lets an AI assistant like Claude connect directly to external tools and APIs, rather than waiting for a human to paste in data. Instead of exporting a CSV and prompting an AI to “analyse this”, the model queries the API itself, in real time, with full parameter control. For keyword work that matters because:

• No loss in translation. You query the DataForSEO API directly — same volume, competition and CPC figures, no manual reshaping.
• Up to 100 keywords per request. Volume, competition index, CPC and a 12-month trend for the whole set in a single call.
• Parallel metrics. Everything arrives together, not one tab at a time.
• Structured output. The model receives JSON it can sort, filter and group — not a screenshot it has to read.

The call is a single tool inside Claude’s session: seo_keyword_metrics(["seo consulting reutlingen", "webdesign reutlingen"]).

03The stack — built on open source

It runs on a standard Linux machine. No cloud infrastructure, no SaaS subscription, no proprietary software — just open components wired together.

04Live session data — Germany region

A real extract from a keyword analysis for the Reutlingen area, run on 18 June 2026 through the MCP server. Competition is the DataForSEO index out of 100; lower is easier to rank.

05Not just volume — trend direction

Beyond static monthly numbers, the server returns Google Trends rising queries — where demand is forming before it becomes obvious. From the same session:

City-specific searches are exploding — a pattern that maps across every German mid-size city. Build a landing page for “seo beratung reutlingen” now and you arrive at exactly the right time.

06A complete SEO intelligence ecosystem

The keyword server is one module of eight — each addresses a different SEO dimension, all wired into Claude Code.

07How we use it in client work

Not a proof of concept — it runs daily in our SEO audit workflow for clients in Reutlingen, Stuttgart and across Germany. A typical loop:

1. Define the target keyword set for the client project.
2. MCP Keywords — volume, CPC and competition for 50–100 keywords in under 60 seconds.
3. MCP SERP — the top-10 competitors per keyword.
4. MCP Authority — whether the client domain can realistically rank.
5. MCP GSC — current visibility and striking-distance keywords.
6. MCP Trends — seasonality and rising queries.
7. Output: a complete keyword brief in a single Claude Code session, zero tool-switching.

08Why this workflow matters

The 2025–2026 search landscape rewards specificity. Generic keyword exports miss the patterns that actually move performance. This loop matters because:

1. It uses your data, not industry benchmarks. The high-CPC, low-competition terms and the rising queries are specific to your market and moment.
2. It closes the loop between research and action. Metrics, assessment and a shortlist land in the same session as the analysis.
3. No SaaS lock-in. One-time setup, full ownership, fractions of a cent per keyword.
4. It surfaces demand early. Rising-query data points at the keywords worth owning before the volume shows up.