</>MCP Agents Market
MCP Server

mnemosyne

by mnemosyne-oss3kPythonUpdated 2026-09-03

Zero-cloud AI memory that works everywhere. SQLite-backed. One pure-Python dependency.

Claude CodeCursorWindsurfCodex

Mnemosyne is a local-first AI memory layer that runs entirely on SQLite without requiring cloud services or external dependencies. It implements a three-tier BEAM (Bilevel Episodic-Associative Memory) architecture combining working memory, episodic long-term storage, and temporal knowledge graphs. The system works with any MCP-compatible client (Claude Code, Cursor, Windsurf, Codex) and agent framework through built-in MCP server support, Python SDK, or platform-specific integrations. Developers get hybrid search combining vector similarity, full-text ranking, and importance scoring, all executed inside a single SQLite database with binary-compressed embeddings for minimal storage overhead.

Key Features

MCP server with stdio, SSE, and streamable HTTP transports for integration with Claude Code, Cursor, Windsurf, and any MCP client
BEAM three-tier memory architecture: working memory for hot context, episodic memory for long-term storage, and TripleStore for temporal knowledge graphs
Hybrid recall scoring combining 50% vector similarity, 30% FTS5 full-text search, and 20% importance weighting entirely within SQLite
Binary vector compression (MIB) reducing 384-dimensional embeddings to 48 bytes with 32× storage savings and Hamming-distance retrieval
Optional bidirectional sync between instances with client-side payload encryption (Fernet or PyNaCl SecretBox)
Platform-native integrations for OpenWebUI, OpenClaw, Pi, and Hermes Agent alongside universal MCP and Python SDK support
Local embedding generation via FastEmbed or sentence-transformers with multilingual model options
Memory banks for per-domain isolation, entity extraction, temporal triple versioning, and global scope for cross-session facts

Use Cases

  • 01Equipping Claude Code or Cursor with persistent memory so the AI remembers user preferences, project context, and past conversations across sessions
  • 02Building multi-session agent workflows where project-specific knowledge, decisions, and entity relationships persist beyond individual chat threads
  • 03Synchronizing desktop and VPS agent instances so memories recorded during local development are available to remote deployed agents
  • 04Enabling agents to maintain temporal knowledge graphs tracking who worked on what and when, with version chains for evolving facts
  • 05Privacy-conscious deployments requiring zero cloud dependencies and optional client-side encryption for synced memory
  • 06Benchmarking memory system performance against LongMemEval and BEAM datasets with reproducible local SQLite storage

Related MCP Servers

View more

markitdown

Python tool for converting files and office documents to Markdown.

MCP Server187.9k

firecrawl

The context API to search, scrape, and interact with the web at scale. 🔥

MCP Server187.6k

prompts.chat

f.k.a. Awesome ChatGPT Prompts. Share, discover, and collect prompts from the community. Free and open source — self-host for your organization with complete privacy.

MCP Server171.8k

langflow

Langflow is a powerful tool for building and deploying AI-powered agents and workflows.

MCP Server155.4k

mnemosyne — FAQ

What is Mnemosyne?+

Mnemosyne is a local-first AI memory system built on SQLite that provides persistent memory storage and retrieval for AI agents and assistants. It runs entirely on your machine with no cloud services required and integrates via MCP, Python SDK, or platform-specific adapters.

How do I add Mnemosyne to Claude Code or Cursor?+

Install with `pip install mnemosyne-memory`, then add an MCP server entry to your client's configuration JSON (claude.json for Claude Code, .cursor/mcp.json for Cursor) with command "mnemosyne" and args ["mcp"]. Restart the client to activate the memory tools.

Which AI clients and frameworks does Mnemosyne support?+

Mnemosyne works with any MCP-compatible client including Claude Code, Cursor, Codex, and Windsurf. It also has native integrations for OpenWebUI, OpenClaw, Pi, Hermes Agent, and any Python-based agent via direct SDK import.

Do I need API keys or external services?+

No external services are required. By default Mnemosyne uses local FastEmbed embeddings with no API keys. You can optionally configure a remote embedding API by setting MNEMOSYNE_EMBEDDING_API_URL and MNEMOSYNE_EMBEDDING_API_KEY for lower-resource environments.

Is Mnemosyne free and open source?+

Yes, Mnemosyne is released under the MIT license and is completely free. All core functionality runs locally with a single pip install and one SQLite file.

How much RAM and storage does Mnemosyne need?+

The core package uses ~50 MB RAM. Adding local embeddings requires ~800 MB, and the full feature set needs ~1.5 GB. Storage is minimal: 10 million messages compress to 7.2 MB in the benchmark tests due to episodic compression and binary vectors.

How do I install mnemosyne?+

Open the source repository on GitHub and follow its README. mnemosyne is a mcp server — MCP Agents Market links you directly to the official repo.

Is mnemosyne free?+

mnemosyne is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.

Related searches

Knowledge & Memory MCP ServersDeveloper Tools MCP ServersBest SkillsBest PluginsBest AgentsTop rated listingsClaude Code toolsBrowse all categories