memanto-mcp

Memanto is a memory agent for AI systems that keeps information available across sessions. It is described as persistent memory for Claude Code, Cursor, Codex, and 14+ other agents, and as a companion memory agent that helps agents focus while the user keeps ownership of what they learn. The product is built on what it calls an information-theoretic search engine.

Its core functions include instant ingestion, conflict resolution, verifiable sources, deterministic search, temporal queries, freshness prioritization, and semantic categorization into 13 types. The page says memories become searchable immediately after they are written and reports under 90 ms recall latency. It also says the system supports built-in RAG, confidence scoring, local embeddings, autonomous categorization, and daily summaries. A comparison section states that it supports remember, recall, and answer, and that it returns relevant results rather than dumping everything into context.

Memanto is presented as working with agents and frameworks across a broad stack. Named integrations include Claude Code, Cursor, Codex, GitHub Copilot, Gemini, OpenAI Codex CLI, Cline, Windsurf, Continue, opencode, Goose, Roo Code, Augment Code, Hermes Agent, CrewAI, LangChain, LangGraph, LlamaIndex, and n8n. The page also mentions REST and MCP, and it shows a CLI with commands such as memanto connect claude-code and memanto ui. A local interactive dashboard is available for managing agents and memories, viewing conflicts and connections, and migrating from Mem0, Letta, and more.

The product is 100% free, open source, and runs entirely on the user’s machine. Installation is shown through pip install memanto, with an on-prem backend using Docker and localhost:8080, and a cloud option that uses an API key. The page also says embeddings and answers can run via local Ollama models, and that no API keys, vector database, or backend service are required for the on-machine setup.