memorius MCP server
Commit Nov 13, 2026Provides a hierarchical memory vault with semantic search and knowledge graphs for AI agents via MCP and REST servers.
Also known as Dream-Pixels-Forge/memorius
- Stars
- 1
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Nov 13, 2026
- License
- MIT
What is memorius MCP?
memorius provides a hierarchical memory vault with semantic search and knowledge graphs for AI agents via MCP and REST servers.
Summary generated from the project's README
About memorius MCP server
Providing persistent, structured memory for AI agents is what Memorius enables through a hierarchical vault system. Developers can organize information using a Vault, Shelf, Folder, and Note structure rather than a flat storage model. It works locally using ChromaDB or SQLite-vec backends, ensuring data stays on your machine without requiring cloud services or external API keys for basic operation.
Users can connect this memory to agents like Claude Code, Codex CLI, or Gemini CLI via a stdio MCP server using the memorius serve command. For more complex setups, a FastAPI REST server can be deployed with TLS and authentication. The system includes advanced features like semantic search with cross-encoder reranking, knowledge graphs for entity relations, and neuroscience-inspired consolidation processes to prune and fact-check stored information.
Beyond standard text retrieval, the tool offers a unique 3D bookcase browser for visual exploration of the vault. It also includes lifecycle hooks for over seven different AI coding agents to automate memory management and security.
What you can do with it
- 1Organize complex knowledge hierarchiesA developer can structure technical documentation into a logical library using vaults, shelves, and folders.
Written from the project's README. Every use case is checked against a quote from it.
README
View on GitHubMemorius
Persistent, searchable memory with hierarchy, semantic retrieval, knowledge graphs, and MCP-native integration — for Claude Code, Codex CLI, Gemini CLI, OpenClaude, OpenCode, Pi, OpenClaw, and any MCP-compatible agent.
📚 Documentation: https://dream-pixels-forge.github.io/memorius/ — CLI reference, MCP server, hooks, REST API, architecture, and the memory engine.
Found this useful? Leave a star ⭐ — it's the best way to support the project.
What Memorius gives you
| 🧠 Hierarchical vault | Vault → Shelf → Folder → Note. Organize memory like a library, not a flat bucket. |
| 🔍 Semantic search | Vector similarity with cross-encoder reranking and temporal-decay ranking. |
| 🕸️ Knowledge graph | Entity co-occurrence graph with path-finding and relation extraction. |
| 🔌 MCP + REST servers | memorius serve (stdio MCP) and memorius serve-rest (FastAPI, TLS, auth). |
| 🪝 Agent hooks | Auto-detecting lifecycle hooks for 7+ AI coding agents — normalizers + security. |
| 📖 3D bookcase | Walkable first-person vault browser (memorius bookcase) with librarian hippo. |
| 🔄 Consolidation | Neuroscience-inspired decay, pruning, and fact-checking to keep memory fresh. |
| 🔒 No lock-in | Local-first, ChromaDB + SQLite-vec backends, no cloud, no required API keys. |
Install
pip install memorius
All features — vector search, cross-encoder reranker, REST API server, SQLite-vec backend, OpenAI embeddings, knowledge graph, agent hooks — are included out of the box. No extra install flags needed.
Verify:
memorius --version
memorius init # creates ~/.memorius/config.yaml + storage dir
Run in a container
The production image is multi-stage, runs as non-root UID 10001, and uses
dumb-init as PID 1, with the vault on a volume at /data.
docker build -t memorius:prod .
docker run --rm -p 127.0.0.1:8912:8912 -v memorius-data:/data memorius:prod
Or with Compose:
docker compose up --build
The REST API is published on loopback only; without MEMORIUS_API_KEY it is
read-only for non-loopback clients. Inside the container the API is plaintext
(--insecure) — terminate TLS in front of it or mount certificates and
override the command. Liveness: GET /health/live.
Quick start
# 1. Initialize a vault
memorius init
# 2. Store a memory
memorius store "The sky is blue because Rayleigh scattering scatters shorter wavelengths more" \
--vault main --shelf science --folder physics
# 3. Semantic search
memorius search "why is the sky blue"
# 4. Mine memories from a conversation
memorius mine transcript.txt --vault conversations
# 5. Check vault status
memorius status
That's the whole loop: store → embed → search. Read the Getting Started guide for the vault hierarchy, agent skill, and a recommended first-week flow.
CLI at a glance
memorius store "text" [--vault/--shelf/--folder/--note/--ttl]
memorius search "query" [--n/--expand-graph/--rerank/--web/--tag]
memorius mine file.txt [--vault/--text]
memorius extract file.txt [--backend openai|ollama|regex]
memorius diary "session" [--title/--summary]
memorius ls [vault]
memorius consolidate [--threshold/--dry-run]
memorius factcheck "stmt" [--web]
memorius profile "session"
memorius prune [--threshold/--dry-run/--delete]
memorius graph [view|export-html|stats|json]
memorius learn "text" --category bug_fix
memorius recall "query"
memorius obsidian [list|import|export]
memorius export ./backup [--format json|markdown]
memorius import ./backup.json [--replace]
memorius doctor # 6 health checks
memorius repair [--reindex/--dry-run]
memorius stats [--json] # vault + memory + graph + doctor summary
memorius status [--json] # vault + embeddings + recent diaries
memorius serve # MCP server (stdio)
memorius serve-rest [--daemon/--stop] # FastAPI server
memorius bookcase # 3D walkable vault browser
Vault Bookcase (3D walkable)
Run the bookcase:
memorius bookcase
It starts the REST API on http://127.0.0.1:8912, serves the bookcase page on http://127.0.0.1:8400/bookcase.html, and opens your browser. Everything is loopback-only — nothing leaves your machine.
Walk: WASD or arrow keys · Shift to hurry · Click to grab mouse-look · / to ask the librarian · M for floor plan · ? for all controls.
Documentation
The full docs live on the site — each concern on its own page:
| Page | What's on it |
|---|---|
| Getting Started | Install, init, first memory, Vault→Shelf→Folder→Note hierarchy, agent skill |
| CLI Reference | Every subcommand and flag (incl. repair and --json) |
| MCP Server | All 29 tools + per-agent client config snippets |
| Agent Hooks | Auto-detecting lifecycle hooks, 10 actions, normalizers, security |
| REST API | Full endpoint table, auth, rate limits, daemon mode, TLS |
| Architecture | Search pipeline, embeddings, vector backends, graph, configuration |
| Agent Compatibility | Which agents, detection priority, how to plug in your own |
| Memory Engine | Decay, consolidation, pruning, fact-checking, reranking, doctor, repair |
License & support
MIT · Full release history in CHANGELOG.md.
Built by Dream-Pixels-Forge. Open source.
memorius MCP questions
How do I use Memorius with an AI agent?
You can use it via the MCP server using the memorius serve command or through its REST API.
Does Memorius require cloud services or API keys?
No, it is a local-first solution that uses ChromaDB and SQLite-vec backends with no required API keys.
How is the memory organized?
Memorius uses a hierarchical structure consisting of Vaults, Shelves, Folders, and Notes.
Can I visualize my memory vault?
Yes, you can use the 'bookcase' command to launch a 3D walkable first-person vault browser.