Yarmoluk/ckg-mcp

Overview

Yarmoluk/ckg-mcp provides 97 domain knowledge graphs for AI agents, accessed via MCP-native traversal. It claims 4× F1 improvement over RAG and 11× fewer tokens per query.

Best for

Best for
Developers building token-efficient AI agents with structured knowledge needs

Use cases

• Optimizing context retrieval for AI agents
• Reducing token usage in knowledge-intensive queries
• Replacing RAG with structured knowledge graphs

How to use

Install

uvx ckg-mcp          # no install — runs immediately

Tools exposed

• Fine-tuning
• RELATES_TO
• CONTRASTS_WITH

Tested with

Claude Desktop, Claude Code, Cursor, Windsurf, Cline, ChatGPT

Notes

Yarmoluk/ckg-mcp provides 97 domain knowledge graphs for AI agents, accessed via MCP-native traversal. It claims 4× F1 improvement over RAG and 11× fewer tokens per query.

3 stars on GitHub. Last updated 2026-07-12. Licensed MIT.

Use cases

• Optimizing context retrieval for AI agents
• Reducing token usage in knowledge-intensive queries
• Replacing RAG with structured knowledge graphs

Pros

• Significant F1 improvement over RAG
• Dramatic token reduction per query
• MCP-native design for agent integration

Cons

• Limited to 97 predefined domains
• Requires MCP-compatible infrastructure
• Low community adoption (3 stars)

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.