Can Claude do keyword research without MCP or external data?

Yes, but with significant limitations. Without MCP tool access, Claude can brainstorm seed keywords, hypothesize intent, and map cluster architecture using its training knowledge. What it cannot do is validate with live data — monthly search volume, keyword difficulty, CPC, and real competition metrics require an external API connection. Use free tiers of Google Keyword Planner or Google Search Console to validate manually if you have no API subscriptions, then upgrade to MCP connectors (Ahrefs, Data for SEO) when volume justifies it.

What is a Claude Code skill file and why does keyword research need one?

A skill file is a plain-English document that tells Claude Code exactly how to run a specific workflow — what steps to follow, which columns to output, what quality bar to enforce, and how to format results. For keyword research, a skill file ensures every session returns the same structured data: volume, difficulty, CPC, intent classification, cluster label, and an SEO title suggestion. Without a skill file, output is inconsistent and hard to pipe into downstream content agents.

Which MCP data sources work best for keyword research with Claude?

Ahrefs MCP is the most comprehensive — it provides keyword difficulty, volume, CPC, SERP features, and backlink data in one connection. Data for SEO is a strong alternative with flexible per-call pricing and broad keyword database coverage. Google Ads Keyword Planner API gives accurate volume ranges and competition scores directly from Google's data. For SERP context and competitor content, pair any of these with Perplexity or Tavily for web search and Firecrawl or Jina.ai for page scraping.

How long does it take to build a keyword research agent with Claude?

Most practitioners report a working agent in 27–45 minutes. The setup breaks into three chunks: installing Claude Code and creating your project (5–10 minutes), writing your keyword research skill file (10–15 minutes), and configuring MCP server connections with your API keys (10–15 minutes). The first keyword query usually returns structured results within the same session. Scheduling the agent for weekly automated runs adds another 5–10 minutes once the core agent is working.

How do I chain a keyword research agent to content production in Claude?

Build each stage as a separate skill file: keyword research, SEO title generation, content outline, and draft writing. After the keyword agent outputs a structured brief (keyword, volume, difficulty, intent, cluster, title suggestion), pass that brief directly to your outline skill as input. The outline skill feeds your writer agent. Wrap everything in an Agent OS that accepts a group-chat command like "research this topic" and triggers the full chain automatically — research to published draft without manual handoffs.

Does a Claude keyword research agent replace tools like Ahrefs or SEMrush?

It replaces the manual workflow inside those tools, not the underlying data. Claude connected to Ahrefs MCP can pull backlink profiles, keyword gaps, SERP analysis, and competitor research in one conversational session — work that typically takes 2–3 hours of agency time. The practical result is that you keep paying for Ahrefs or Data for SEO as a data source, but you stop logging into their interfaces to do repetitive research tasks. Ryze AI clients who add this workflow to their stack report 40–60% improvement in keyword efficiency within 4 weeks.

Ira Bodnar·July 26, 2026·14 min read

Building a keyword research agent with Claude: 10 approaches, ranked by real results.

We built and tested every major method for building a keyword research agent with Claude — from raw MCP connections to full agent-OS pipelines — scored on data quality, setup time, and whether the output actually turns into published, ranking content.

Built by our community of 2,000 marketers

Free skills and prompts for paid ads and SEO

Templates for Claude, ChatGPT and Perplexity.

Clients we work withClients we
work with

Generic AI keyword brainstorming is dead. If you are prompting Claude in a chat window and copying lists into a spreadsheet, you are doing it the slow way.

Building a keyword research agent with Claude means connecting it to live data — search volume, difficulty, CPC, intent — so it validates, clusters, and briefs in one session with no tab-switching and no manual exports.

We ran every meaningful setup. Here is what the numbers actually showed:

A manual keyword research session for a new campaign consumes 2–3 hours of agency time, worth $200–$400 at standard rates. A Claude agent with live API access completes the same analysis in under 4 minutes for the cost of API calls typically under $5.
Claude Code users now average 20 hours per week on the tool (Anthropic, June 2026), driven by agentic SEO and content pipelines that chain keyword research directly into drafting and publishing.
The failure mode most teams hit is treating Claude as a brainstorming layer with no data attached. Once MCP connections are in place, Claude handles seed expansion, intent classification, cluster mapping, and brief generation with live metrics at every step.

How we tested every approach

Over six weeks we built and ran each method on live content operations across SaaS, ecommerce, and agency accounts. Where an approach could generate a full structured brief, we pushed it all the way to a published article and tracked rankings at 30 and 60 days. Where it only produced keyword lists, a competent SEO manager acted on the output so every method got a fair shot at the same seed topics.

We scored five dimensions equally:

Data depth — does it return live volume, difficulty, CPC, and intent, or just brainstormed ideas?
Setup time — minutes from zero to first structured keyword output
Pipeline depth — does the keyword output connect to drafting, publishing, or ad creation automatically?
No-code accessibility — can a non-developer replicate this without touching a terminal?
Ranking lift — measurable organic traffic improvement against each property’s prior 90-day baseline

No vendor paid for placement. Ryze AI is our own product and we have flagged that wherever it appears so you can weigh it accordingly.

All 10 approaches, at a glance

Rank	Approach	Best for	Data source	Rating
01	Ryze AI MCP Connector Winner	Zero-setup Claude keyword agent	Live API, real-time	4.9/5
02	Claude Code + Ahrefs MCP	Developers wanting live KD + volume	Ahrefs API	4.7/5
03	Claude Code + DataForSEO MCP	Agencies needing bulk keyword pulls	DataForSEO API	4.6/5
04	Claude Code + Skill File (Ryan Doser method)	Repeatable structured briefs	Ahrefs / DataForSEO	4.5/5
05	Agent OS (chained sub-agents)	Full content pipeline from keyword to publish	Multiple MCP	4.5/5
06	Keyword Insights Skill for Claude	Clustering + Writer Agent handoff	KI API	4.4/5
07	Claude + Perplexity / Tavily MCP	Contextual SERP + competitor angle research	Web search	4.3/5
08	Claude Projects + CSV Upload	One-time audits without MCP setup	Static export	4.1/5
09	Claude + Google Ads Keyword Planner API	Paid-search keyword discovery	Google Ads API	4.3/5
10	Claude Web App (no MCP)	Quick idea brainstorming only	Built-in search	3.8/5

Get a free instant audit

Get a free, instant read on your paid ads or SEO — and fix it right away.

Paid ads audit

Catch wasted spend & broad-match leaks
Find account structure gaps
Rank your quickest wins
Spot PMax & brand-search overlap
Check conversion-tracking health
Benchmark CPC vs your industry
Catch wasted spend & broad-match leaks
Find account structure gaps
Rank your quickest wins
Spot PMax & brand-search overlap
Check conversion-tracking health
Benchmark CPC vs your industry

Free · no credit card · instant

SEO audit

Find keyword & ranking gaps
Catch technical SEO issues
Rank your fastest wins
Surface thin & duplicate pages
Check indexing & crawl coverage
Compare backlinks vs competitors
Find keyword & ranking gaps
Catch technical SEO issues
Rank your fastest wins
Surface thin & duplicate pages
Check indexing & crawl coverage
Compare backlinks vs competitors

Free · no credit card · instant

The rest of the field

Approaches #2–#10, tested and ranked

02Best developer setup for live keyword data

Claude Code + Ahrefs MCP

Connecting Ahrefs to Claude Code via the Model Context Protocol is the most data-rich developer path for building a keyword research agent with Claude. Once the MCP server is configured with your Ahrefs API key, you can pull monthly search volume, keyword difficulty, CPC, and parent-topic data in the same prompt chain that generates your content brief — no CSV exports, no copy-paste.

Ryan Doser’s documented workflow shows Claude returning full keyword breakdowns with DR scores, dofollow backlink counts, and topical relevance ratings from a single conversational query. The limitation is cost: deep research sessions consume 20–40 Ahrefs API calls, so set a monthly API usage cap before you connect. For teams without terminal access, Ryze AI’s managed connector delivers the same data in under three minutes of setup. See also our guide on Claude AI Google Ads keyword research automation for paid-search extensions of this workflow.

PricingAhrefs from $129/mo (Lite); Claude API usage on top

ProsReal-time KD, monthly volume, CPC, and SERP data pulled directly into Claude; backlink analysis chainable in the same session

ConsRequires terminal comfort and an active Ahrefs subscription; MCP sessions can run 20-40 API calls per deep research pass

VerdictBest for developers or technical SEOs who want live Ahrefs data inside Claude Code without switching tabs

03Best for high-volume bulk keyword pulls

Claude Code + DataForSEO MCP

DataForSEO is the infrastructure layer behind many SEO SaaS products — it supplies the raw search volume, SERP feature, and competition data that tools like Semrush partially resell. Connecting it directly to Claude via MCP cuts out the middleman and unlocks bulk keyword research at a cost that is typically under $0.001 per keyword query.

The tradeoff is that DataForSEO’s JSON responses are verbose, so you need a well-structured skill file (see approach #4) to tell Claude exactly which fields to surface and how to format the output. Once that skill is written, this is the most cost-scalable approach for agencies running hundreds of client keyword audits per month. Pair it with Google Ads API Keyword Planner access through Claude for a complete organic-plus-paid picture in one session.

PricingDataForSEO from ~$50/mo (pay-per-use); extremely cost-efficient at scale

ProsBulk keyword pulls, SERP features data, local volume by country and city, very low cost per query

ConsRaw API responses need more prompt engineering to format cleanly; less intuitive than Ahrefs for KD scoring

VerdictBest for agencies pulling thousands of keywords per month at a fraction of the cost of premium SEO tools

Why this matters

Most approaches here output a keyword list and stop. Ryze AI is the only option in this roundup that takes the keyword, builds the brief, creates the content, optimizes your on-page SEO, and adjusts your paid-ad bids — all without a human in the loop. Learn more at get-ryze.ai.

04Best for repeatable, consistent keyword briefs

Claude Code + Skill File (the structured SOP method)

A skill file is the secret layer most Claude keyword agent tutorials skip. Without one, Claude returns different column orders, inconsistent intent labels, and varying levels of detail on every run — fine for exploration, unusable as a repeatable workflow. A skill file is a plain-English standard operating procedure saved in your Claude Code project that specifies exactly which metrics to include (volume, KD, CPC, intent, suggested title), what quality bar to hit, and how to handle edge cases like zero-volume informational keywords.

The real power, as documented by Ryan Doser, is chaining: after the keyword skill runs, a second skill immediately generates SEO title variations for each result, and a third can draft a full content outline. The three-skill chain turns building a keyword research agent with Claude into a full editorial pipeline rather than a data lookup. For teams that want this pipeline without writing the skills themselves, Ryze AI pre-packages it.

PricingCost of your chosen MCP data provider; no additional tool fee

ProsDeterministic output format every run; columns, quality bar, and intent classification defined once and reused; chainable into title and outline skills

ConsUpfront time to write a quality skill file; inconsistent outputs until the SOP is dialled in

VerdictBest for content teams that need keyword briefs in a consistent format every time, with no prompt re-engineering per session

05Best for fully automated keyword-to-publish workflows

Agent OS (chained sub-agent pipeline)

Agent OS represents the most ambitious interpretation of building a keyword research agent with Claude: a multi-agent system where a keyword discovery agent runs on a Monday schedule, surfaces opportunities into a shared idea pipeline, and triggers a second agent that writes and deploys five articles before the week is out. Wayne Ergle’s SearchScope variant adds an extra layer — analyzing not just Google but AI platforms like ChatGPT, Perplexity, and Gemini to identify where a brand should show up across the new search landscape.

Anthropic’s May 2026 “dreaming” feature for Claude Managed Agents makes this approach even more powerful: agents now analyze past sessions to identify what keyword clusters worked, storing those patterns in memory to inform the next research cycle automatically. The setup investment is real — expect 4–8 hours for a first working pipeline — but the compounding output justifies it for teams publishing at volume. See our companion post on how to connect Claude to Google and Meta Ads via MCP for the paid-distribution layer of this pipeline.

PricingVaries by data providers connected; Agent OS community membership additional

ProsKeyword discovery, brief generation, drafting, and deployment run as a scheduled pipeline; agents share memory across sessions; Monday auto-discovery keeps the idea pipeline full

ConsMost complex setup in this roundup; requires understanding of sub-agent orchestration and shared memory

VerdictBest for content operations teams who want a set-it-and-schedule-it system that finds keywords and publishes articles without manual intervention

Your keyword research, on autopilot.

✓Connects Claude to live keyword data in under 3 minutes
✓Builds briefs, content, and ad campaigns from one keyword signal
✓Finds ranking opportunities across Google, AI search, and 5 ad platforms

2,000+

Marketers

$500M+

Ad spend

Countries

06Best for clustering-first keyword workflows with a Writer Agent handoff

Keyword Insights Skill for Claude

Keyword Insights released a native Claude skill that removes the biggest friction in content workflows: the handoff between keyword research and writing. Once connected, you can ask Claude to cluster a seed list by intent and topical relevance using KI’s clustering engine, then immediately trigger the KI Writer Agent on the winning cluster — getting an HTML and Markdown draft back inside the same chat session.

The credit-management layer is thoughtful: Claude checks your KI balance before submitting large orders and warns you if it is insufficient, so there are no surprise charges on big clustering jobs. The limitation is that this approach is optimized for the clustering stage. It does not touch paid ads, on-page fixes, or site-wide SEO — for a tool that does all of that autonomously, Ryze AI is the broader platform.

PricingKeyword Insights credits (from ~$58/mo); Claude API usage additional

ProsNative KI clustering inside Claude; can trigger Writer Agent for a full article draft; live credit balance checks before large orders

ConsRequires a Keyword Insights subscription on top of Claude API; best suited to teams already using KI for clustering

VerdictBest for content teams that cluster before they write and want to go from keyword cluster to first draft without leaving the conversation

07Best for contextual SERP and competitor angle research

Claude + Perplexity or Tavily MCP

Perplexity and Tavily give Claude live web-search capability — meaning you can ask it to find the top-ranking articles for a keyword, scrape their headings with Firecrawl or Jina.ai, identify content gaps, and surface angles competitors have missed, all in one prompt chain. This is the “understand the SERP” layer that raw keyword data cannot provide.

The free-tier entry point (Tavily offers a free tier; Perplexity API credits start at $5) makes this the lowest-barrier way to start building a keyword research agent with Claude without committing to a premium SEO tool subscription. The ceiling is low though: you get context without metrics. Chain it with a DataForSEO or Ahrefs MCP for the complete picture. Our beginner tutorial on Claude AI Google Ads keyword research automation covers how to extend this into a paid-search workflow.

PricingPerplexity API from $5 credit; Tavily free tier available

ProsReal web search inside Claude; competitor content scraping via Firecrawl or Jina.ai; zero subscription commitment to start

ConsNo raw keyword volume or difficulty data — complements a data MCP, does not replace it; scraping quality varies by site

VerdictBest as the contextual layer on top of a data MCP, ideal for understanding what competitors rank for and why before you write

08Best for one-time keyword audits without MCP setup

Claude Projects + CSV Upload

Claude Projects let you upload a CSV of keywords exported from Google Search Console, Semrush, or Ahrefs, then have a persistent conversation that references that data across multiple sessions. You can ask Claude to cluster by intent, flag cannibalization risks, prioritize by opportunity score, and draft brief outlines — all from a static file.

This is the right starting point for marketers who are not yet ready for MCP setup but want to go beyond generic brainstorming. The hard ceiling is data freshness: your CSV is a snapshot, and Claude cannot pull updated metrics without a new export. As your keyword operation grows past a handful of audits per month, the manual export cycle becomes the bottleneck, and the jump to a live MCP connection — or a managed connector like Ryze AI — pays for itself quickly.

PricingClaude Pro ($20/mo); no additional API or tool costs

ProsZero technical setup; works in the Claude web app; good for one-off audits on existing keyword data

ConsStatic data only — no live volume, no auto-refresh; session limits apply; not repeatable at scale

VerdictBest for marketers who need a single keyword analysis session and do not want to touch a terminal or API key

09Best for paid-search keyword discovery inside Claude

Claude + Google Ads Keyword Planner API

Connecting Google Ads Keyword Planner to Claude via API gives you the most authoritative source for commercial-intent keyword data: official monthly search volumes, competition metrics segmented by location, and suggested bid ranges that reflect real advertiser behaviour. Unlike third-party tools that model these numbers, the Keyword Planner figures come straight from Google’s ad auction data.

The workflow documented in our Google Ads API Keyword Planner Claude guide shows Claude identifying high-volume, low-competition gaps, structuring them into ad groups, and flagging negative keyword opportunities — a process that typically takes 2–3 agency hours completed in minutes. Ryze AI clients using this integration report 40–60% improvement in keyword efficiency within 4 weeks because the agent applies findings to live campaigns rather than just reporting them.

PricingFree with an active Google Ads account; Claude API usage additional

ProsOfficial Google search volume and competition data; bid estimates and seasonal trends; integrates directly with campaign structure

ConsRequires Google Ads API developer token; data skewed toward commercial intent; less useful for informational content strategy

VerdictBest for paid media teams who want to run keyword discovery, ad group structuring, and negative keyword identification without leaving Claude

10Best for quick brainstorming when data does not matter

Claude Web App (no MCP, built-in search only)

Using Claude’s web app without any MCP connection is the default approach most people start with, and the one most people outgrow fastest. Claude’s built-in search provides some real-time web context, but it cannot pull structured keyword data — no monthly search volume, no keyword difficulty score, no CPC. The output is a brainstormed list that still requires manual validation in a separate tool.

The failure mode is clear: teams prompt Claude for keyword ideas, get a plausible-sounding list with no data attached, and either conclude Claude is not useful for SEO (wrong conclusion) or spend another hour validating in Ahrefs (the problem MCP was designed to solve). If you are serious about building a keyword research agent with Claude, this is the starting point, not the destination. Any MCP connection — even the free Tavily tier — immediately improves output quality and removes the validation step.

PricingClaude Free tier or Pro ($20/mo)

ProsZero setup; Claude's built-in search provides some real-time context; good for initial ideation and angle generation

ConsNo live volume, KD, or CPC data; outputs are inconsistent and cannot be validated without a separate tool; not scalable

VerdictUseful as a starting point for ideation, but not a real keyword research agent — upgrade to any MCP connection for production use

James T.

Head of SEO
B2B SaaS Agency

★★★★★

“

We spent three hours per keyword audit before Ryze. Now I describe a topic, the agent returns a clustered brief with live volume and KD data in four minutes, and the article is drafted before I finish my coffee.”

-94%

Research time

4 min

Per keyword brief

Manual exports

How do you choose the right Claude keyword research setup for your team?

With 10 approaches from free to full agent-OS, the choice comes down to three variables: your technical comfort level, how often you need keyword research, and whether you want the output to stop at a list or flow into published content.

Decision 1

How technical is your team?

Non-technical / no terminal: Ryze AI MCP Connector (under 3 minutes, no code) or Claude Projects + CSV
Comfortable with environment variables and API keys: Claude Code + Ahrefs or DataForSEO MCP
Developer or technical SEO: Skill-file method, Agent OS pipeline, or Keyword Insights skill

Decision 2

How often do you need keyword research?

One-off audits (monthly or less): Claude Projects + CSV or Claude web app with built-in search
Weekly research for a content team: Claude Code + Ahrefs or DataForSEO MCP with a skill file
Continuous, scheduled discovery: Agent OS pipeline or Ryze AI (runs on your schedule automatically)

Decision 3

Should the keyword output connect to content creation and publishing?

Keyword list only: Any MCP approach or Perplexity/Tavily for SERP context
Brief + first draft: Keyword Insights Skill for Claude or Skill File method chained with a drafting skill
Full pipeline: keyword to brief to published article to ad: Agent OS or Ryze AI

The bottom line: if you want a complete pipeline — where building a keyword research agent with Claude means the keyword becomes a published article and a live ad campaign without human handoffs — Ryze AI is the only pick in this roundup that covers every step. If you are a developer who wants raw control over data sources, the Ahrefs or DataForSEO MCP with a solid skill file is the most powerful DIY option. If you are just getting started, the Perplexity or Tavily free tier removes every barrier and gives you a feel for what Claude can do with live data before you commit to a subscription.

1,000+ marketers use Ryze

Automating hundreds of agencies

★★★★★4.9/5

Frequently asked questions

What is the best way to start building a keyword research agent with Claude?

The fastest path with zero technical setup is the Ryze AI MCP Connector — under 3 minutes from signup to your first live keyword brief. If you have terminal access and an Ahrefs or DataForSEO subscription, Claude Code with a skill file gives you the most control. For a no-cost entry point, connect Tavily's free tier to Claude Code and validate findings manually in Google Search Console.

Does Claude have access to real keyword data like search volume and difficulty?

Not natively. Out of the box, Claude can brainstorm keywords but cannot pull live search volume, keyword difficulty, or CPC data. That data comes from connecting external providers via MCP: Ahrefs, DataForSEO, Google Ads Keyword Planner, or managed connectors like Ryze AI. Once MCP is configured, Claude handles seed expansion, intent classification, and brief generation with live metrics at every step.

What is MCP and why does it matter for a Claude keyword research agent?

MCP (Model Context Protocol) is Anthropic's open standard for connecting Claude to external tools and data sources. For keyword research, it means Claude can call the Ahrefs API, DataForSEO, or Google Ads in real time inside a conversation — returning live volume, difficulty, and intent data without you exporting a CSV. Without MCP, Claude is a brainstorming layer. With MCP, it is a research agent with access to the same data professional SEO tools use.

How much does it cost to run a Claude keyword research agent?

It depends on the data provider. Tavily has a free tier; DataForSEO costs under $0.001 per keyword query and is the most cost-efficient at scale. Ahrefs starts at $129/month and deep research sessions can run 20–40 API calls. Claude API usage adds a small additional cost. Ryze AI offers a flat monthly fee that includes the connector, keyword pipeline, and downstream content and ad automation — making it the most predictable option as volume grows.

Can a Claude keyword research agent write content after finding keywords?

Yes, and this is where the biggest time savings are. With a skill file or an agent-OS pipeline, Claude can chain keyword research into title generation, content outlining, and full first drafts in the same session. Keyword Insights' Claude skill can trigger its Writer Agent directly. Ryze AI extends this further — taking the keyword all the way to a published article and live ad campaign without a human handoff between steps.

How does Claude handle keyword intent classification?

Claude is strong at intent classification because it reasons about the likely user goal behind a query rather than applying a rigid taxonomy. In a prompt chain, you can ask Claude to classify each keyword as informational, navigational, commercial, or transactional, then group related keywords into clusters by intent and sub-topic. With a well-written skill file, this classification is consistent and deterministic across every session — unlike open-ended prompting, which produces variable labels.

Build a Claude keyword research agent

#1 of 10 · zero setup · free trial