Kremis MCP server

Kremis is a small helper program that gives your AI assistant a memory of its own. You feed it facts, and later your AI can look those facts up instead of guessing. It is handy if you want answers you can trace back to real data, not made-up ones.

What this MCP server does

You ask your AI a question about your data. The AI asks Kremis through this helper. Kremis looks in its local graph of stored facts and returns what it finds. If the fact is not there, it says so plainly instead of guessing. You get the answer in the chat, and it traces back to a real stored signal.

Click to zoom

What you can do with it

• Store facts as nodes in a local graph
• Look up a fact by its entity
• Find the strongest path between two entities
• Check whether a claim is supported by stored data
• Get an explicit not found when data is missing
• Export the whole graph as a snapshot for auditing
• Verify the graph state with a hash

Try asking your AI

• “Does service A depend on service B, directly or through other services?”
• “Look up everything we know about entity 42.”
• “What is the strongest path between these two items in my graph?”
• “Is this claim supported by what I stored, or is it unknown?”

What it gives back to you

You get plain answers in the chat: the fact, the path, or a clear not found. Each answer carries a grounding label so you can see whether it is a stored fact, an inference, or unknown. You can also ask for a certificate that proves a fact, or proves its absence, tied to a hash of the graph.

Before you start

Who it's for

People who need answers they can trace back to real data, like auditors, analysts, and anyone using an AI assistant for careful fact-checking.