Model Context Protocol
Your AI forgets you
the moment the session ends.
Add Memcone as an MCP server in one command. Your editor gets persistent memory with the same three primitives: context, remember, and recall.
One command. Your AI remembers everything from here on.
Get started
Exposed tools
memcone.contextRetrieve the most relevant memory for this scope as a ready-to-inject string. Called before each response — one call, no prompt engineering required.
Before every response
memcone.rememberStore a fact or event in memory. Extracts, deduplicates, and embeds automatically. Contradictions resolve without manual cleanup.
When user reveals something
memcone.recallSemantic search over stored memory. Returns ranked results matching a specific query.
When user references the past
How it works
IDE fetches tool manifest
Your agent runtime reads /.well-known/mcp.json and registers the three Memcone tools automatically.
Model decides when to call
The LLM reads the tool descriptions and decides autonomously when to fetch context, store facts, or recall memory — you don't prompt it.
Memcone handles everything else
Storage, embedding, caching — all transparent to the model. It just gets clean memory text back.