Modern SEO requires moving from flat keyword lists to semantic topical authority graphs and grounded Retrieval-Augmented Generation (RAG).
The seosiri-keyword-cluster-mcp server provides a local-first Model Context Protocol gateway to group keywords into semantic clusters, classify search intent, flag cannibalization conflicts, and run an in-memory RAG vector search engine to ground AI responses in verified site content.
A deep technical guide to deploying a local-first keyword clustering, search intent classification, and vector RAG retrieval server using Python, FastMCP, and PyPI.
Lead Architect: Momenul Ahmad | Organization: SEOSiri-Official | Package Registry: PyPI (seosiri-keyword-cluster-mcp v1.0.1) | Master Hub: SEOSiri Central Directory
1. The Shift to Semantic Keyword Clustering and Grounded RAG
In modern search engine and AI agent discovery, targeting un-clustered individual keywords leads to content duplication and poor topical authority. AI models and search engines evaluate how comprehensively a domain covers a core topic across interconnected pillar and cluster nodes:
- Semantic Token Overlap: Grouping search queries by shared n-gram overlap and token similarity prevents creating redundant pages for near-identical search intents.
- Keyword Cannibalization Risks: Mapping multiple URLs to overlapping queries confuses search crawlers and dilutes domain authority.
- Grounding via Vector RAG: AI agents assisting with content generation or technical audits require retrieval-augmented context from verified site passages to prevent hallucinations.
To address these operational needs, SEOSiri engineered seosiri-keyword-cluster-mcp—a local-first Model Context Protocol server that equips AI clients with deterministic tools for keyword clustering, intent classification, and vector-based RAG context retrieval.
2. The Local-First In-Memory RAG Vector Engine
The server embeds a lightweight, local-first Vector RAG Engine built on in-memory SQLite and deterministic term-frequency embedding math. This allows indexing article passages and retrieving grounded context without external vector database subscriptions:
- 384-Dimensional Normalized Vectors: Text chunks are converted into L2-normalized 384-dimensional feature vectors in memory.
- In-Memory SQLite Vector Store: Passage chunks, source URLs, and JSON vector strings are stored in an active in-memory database table.
- Cosine Similarity Search: User prompts are embedded and evaluated against stored vectors using dot product similarity math, returning the top-$K$ most relevant grounded passages.
3. Core Tool Inventory (13 Production Tools)
The seosiri-keyword-cluster-mcp server exposes 13 production-tested tools to connected AI agents:
- cluster_keywords_by_similarity: Groups keyword lists into semantic clusters using token overlap.
- classify_search_intent: Categorizes search queries into Informational, Navigational, Commercial, or Transactional.
- detect_keyword_cannibalization: Identifies conflicting URLs targeting identical search queries.
- generate_topical_authority_map: Compiles pillar-cluster content taxonomy trees.
- calculate_keyword_difficulty_score: Computes algorithmic query difficulty scores ($0\text{--}100$).
- extract_lsi_semantic_variants: Identifies Latent Semantic Indexing terms and entity phrases.
- generate_text_embeddings: Generates normalized 384-dimensional vector embeddings for content chunks.
- index_content_chunks_for_rag: Stores text chunks, vector embeddings, and URL metadata into the local RAG vector store.
- retrieve_semantic_rag_context: Performs cosine similarity search across indexed corpus chunks to retrieve top-$K$ grounded context.
- export_cluster_parquet_buffer: Formats keyword clusters into columnar Parquet buffers for DuckDB/S3.
- sanitize_keyword_payload: Strips scripts and malformed tags from inputs.
- get_live_keyword_throughput_metrics: Monitors system health and processing latency.
- get_keyword_server_specifications: Returns protocol specs and capability matrices.
To inspect source code, review commits, or contribute, visit the official open-source repositories on GitHub
4. Synergy Across the SEOSiri Intelligence Suite
In enterprise SEO and AI search optimization workflows, seosiri-keyword-cluster-mcp integrates with our other specialized intelligence engines:
- Content Schema Server (content-schema-mcp): Generates multi-entity TechArticle/FAQPage JSON-LD schemas, validates GA4 report metrics, and calculates content stickiness.
- AEO/GEO Intelligence Server (aeo-geo-mcp): Evaluates AI readiness scores, audits /llm.txt compliance, and extracts direct-answer cards for Perplexity and SearchGPT.
- DNS SEC Audit Server (dns-sec-audit-mcp): Audits A/AAAA/MX records, SOA Expire timers, SSL/TLS certificates, and HTTP security headers.
- ETL Data Pipeline (etl-pipeline-mcp): Handles multi-source webhook ingestion, SHA-256 PII scrubbing, identity stitching, and data warehouse exports.
5. Installation & Client Configuration Options
Developers and SEO teams can deploy the seosiri-keyword-cluster-mcp package using any of the following installation methods:
Option A: Direct PyPI Installation
Install the latest stable release directly from PyPI:
pip install seosiri-keyword-cluster-mcp
Option B: Zero-Setup Execution via uv (Claude Desktop / Cursor)
To run the MCP server directly from GitHub without installing global dependencies, add this to your claude_desktop_config.json:
{
"mcpServers": {
"seosiri-keyword-cluster": {
"command": "uv",
"args": [
"run",
"--github",
"SEOSiri-Official/keyword-cluster-mcp",
"src/main_server.py"
]
}
}
}
Cloudflare Edge Gateway: You can also route requests through our live Cloudflare Worker gateway at keywords.seosiri.com.
6. Commercial B2B Solutions & Engineering Consulting
Beyond our open-source tools, SEOSiri offers technical consulting, custom RAG vector database architecture, and dedicated deployment services for enterprise organizations and digital agencies:
- Custom Keyword & Content Taxonomy Design: Building automated pillar-cluster mapping pipelines for enterprise e-commerce and media platforms.
- Enterprise RAG Vector Store Integration: Connecting local-first vector search engines to internal knowledge bases and product catalogs for grounded AI search.
- Custom MCP Server Engineering: Developing bespoke Model Context Protocol servers to connect proprietary APIs and databases directly to AI agents.
To discuss custom pipeline engineering or B2B consulting, reach out directly:
- Ecosystem Central Directory: seosiri.com/2026/07/seosiri-mcp-servers.html
- Enterprise Support Email: info@seosiri.com
Executive Summary
The SEOSiri Keyword Cluster MCP (seosiri-keyword-cluster-mcp) provides a local-first technical engineering engine for semantic keyword intelligence and vector RAG retrieval:
- Semantic Clustering: Groups keyword lists into structured pillar-cluster taxonomy nodes using token overlap.
- Search Intent Engine: Automatically classifies queries into Informational, Navigational, Commercial, and Transactional classes.
- Local-First RAG Retrieval: Embeds text into 384-dimensional vectors and runs cosine similarity search to ground AI answers in verified content.
- Cloudflare Edge Gateway: Deployed via Cloudflare Workers (keywords.seosiri.com) for low-latency global routing.
Query Answers
What is seosiri-keyword-cluster-mcp?
seosiri-keyword-cluster-mcp is an open-source Model Context Protocol server developed by SEOSiri to automate semantic keyword clustering, search intent classification, cannibalization detection, and local-first RAG vector retrieval.
How does the built-in RAG engine prevent AI hallucinations?
The server converts content chunks into 384-dimensional vector embeddings, stores them in an in-memory SQLite store, and executes cosine similarity searches to retrieve grounded, factual context passages for AI models.
Which search intent types are classified?
The server automatically categorizes queries into Informational, Navigational, Commercial, and Transactional intent classes.
