ckg-mcp · Pro tier

One install.
Every domain.
Any LLM.

Works with Claude, GPT-4o, Gemini, Llama, Mistral — and any tool that speaks MCP. Set one env var. Unlock everything.

4×
F1 vs RAG
11×
Fewer tokens
97
Domains (Pro)
0
Hallucinated edges
Free
$0
forever
pip install ckg-mcp →
Most popular
Pro
$99/mo
billed monthly · cancel anytime
Upgrade to Pro — $99/mo →
After you pay — 60 seconds to activate.
1
Check your email — your license key arrives immediately after checkout. It looks like xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx.
2
Set the env var — in your shell, Claude Desktop config, or .env: 
export CKG_API_KEY=your-license-key-here
3
That's it. Restart your MCP client and run list_domains — all 97 domains appear. No reinstall, no version change. 
list_domains()
# → Available domains (97): agent-reliability, ai-governance,
#   algebra-1, bioinformatics, blockchain, calculus,
#   context-as-a-service, databricks, hipaa-compliance,
#   icd-10, nvidia-gpu-inference, snowflake, ...
How the key works: After checkout, Stripe redirects to a confirmation page with setup instructions — your key arrives by email within 60 seconds. Set CKG_API_KEY in your MCP config or shell → restart your client → list_domains() shows all 97. No reinstall, no version change. The same binary, unlocked.
Works with every LLM and agent framework.

MCP (Model Context Protocol) is an open standard — ckg-mcp plugs into any client that supports it. One pip install, one env var, any model.

Claude Desktop + API GPT-4o / ChatGPT Gemini Llama 3 Mistral DeepSeek LangChain LangGraph smolagents CrewAI Cursor Windsurf Any MCP client
Why it's worth $99/month.
MetricRAGCKG (this tool)Improvement
Macro-F1 accuracy0.1230.4714× more accurate
Tokens / query2,98226911× fewer tokens
Cost / correct answer$0.0106$0.0010~10× cheaper
5-hop reasoning F10.1700.7724.5× on hard queries
Full run cost (7,928 q)$76.23$7.81~10× cheaper
Fabricated edgesvariable0 — by constructionauditable by design

Benchmark v0.6.2 · 45 domains · 7,928 queries · open methodology · read the paper →

Also independently validated: arXiv:2603.14045 (U. Victoria / Santa Clara) finds 73–84% of GraphRAG errors are reasoning failures — the exact problem CKGs solve by construction. The graph doesn't guess — it traverses.