Ckg Mcp
Give Claude Code or Cursor structured domain knowledge via MCP instead of chunky RAG so answers stay accurate and token-cheap while you ship.
Overview
ckg-mcp is a MCP server for the Build phase that serves structured domain knowledge to agents with far lower token use than typical RAG.
What is this MCP server?
- Structured domain knowledge graph exposed over MCP stdio (PyPI package ckg-mcp v0.1.1)
- Positioned as 42x more accurate than RAG and 11x fewer tokens in official description
- stdio transport for local agent hosts (Claude Desktop, Claude Code, compatible MCP clients)
- Open-source on GitHub (Yarmoluk/ckg-mcp) for self-hosted knowledge bases
- Registry version 0.1.1 on PyPI identifier ckg-mcp
- Marketing claims: 42x accuracy vs RAG, 11x fewer tokens
- Transport: stdio
Community signal: 3 GitHub stars.
What problem does it solve?
RAG over messy docs wastes tokens and still misleads the agent when you need dependable domain facts while coding.
Who is it for?
Indie builders running Claude Code or Cursor who maintain a real domain model and want MCP-native knowledge instead of vector search.
Skip if: Teams that only need occasional doc search with no graph curation, or hosts that cannot run stdio MCP servers.
What do I get? / Deliverables
Your agent queries a structured knowledge graph over MCP so answers align with your domain and context stays lean.
- stdio MCP connection the agent uses for structured knowledge queries
- Lower-token domain answers versus typical RAG setups per project claims
Recommended MCP Servers
Journey fit
Agents need reliable context during build; CKG is the canonical shelf for agent-tooling knowledge plumbing. Fits agent-tooling because it exposes structured knowledge tools to the coding agent, not a one-off frontend widget.
How it compares
MCP knowledge-graph server, not a vector-RAG skill or generic web-search plugin.
Common Questions / FAQ
Who is ckg-mcp for?
Solo and small-team builders using MCP-enabled agents who want structured, accurate domain knowledge without heavy RAG pipelines.
When should I use ckg-mcp?
When you are in Build or Iterate and the agent repeatedly gets domain details wrong or burns tokens on large retrieved chunks.
How do I add ckg-mcp to my agent?
Install the ckg-mcp PyPI package, configure it as a stdio MCP server in Claude Code or your client, and point it at your curated knowledge graph.