mcp-ariel-memory
io.github.Cipher208/ariel-memory
Overview
Universal two-layer memory MCP server for AI agents with stdio and Streamable HTTP transports, 37 tools, RAG, knowledge graphs, wiki, auth, backups, dashboard, and metrics.
Documentation
Installation
Recommended MCP client launch via npm:
npx mcp-ariel-memory --transport stdio
The npm wrapper requires Node.js >=18 and Python 3.10+; its postinstall installs the Python package. The repository also documents pip, Docker, and source installs:
pip install git+https://github.com/Cipher208/mcp-ariel-memory.git
python -m mcp_server --transport stdio
docker build -t ariel-memory .
docker run -p 8000:8000 ariel-memory
git clone https://github.com/Cipher208/mcp-ariel-memory.git
cd mcp-ariel-memory
pip install -e ".[all]"
python -m mcp_server --transport stdio
Configuration
Default stdio launch uses npx mcp-ariel-memory --transport stdio. HTTP launch is documented with python -m mcp_server --transport http --port 8000; dashboard and metrics are enabled with --dashboard. Runtime storage defaults to ~/.mcp-ariel-memory and can be changed with MCP_MEMORY_DATA_DIR. Dashboard/API bearer auth can use MCP_AUTH_TOKEN; otherwise the server loads or creates a token file.
Optional CLI flags: --transport (stdio or http, default stdio), --host (default 0.0.0.0), --port (default 8000), and --dashboard (default false).
Capabilities
The docs describe 37 MCP tools covering user and agent memory CRUD, sessions, episodes, graph operations, auth API key management, backup, saga, read-only replica sync, import/export, cleanup, emergency purge, context injection, and hybrid RRF search. Features include FTS5 plus vector/hashing fallback search, epistemic and temporal graphs, wiki folders, hooks, audit, rate limiting, metrics, automatic backups, and dashboard endpoints.
Limitations and operational notes
Requires Python 3.10+ and pip; npm wrapper requires Node.js >=18. SQLite data is local by default. HTTP dashboard/API endpoints are protected by bearer auth and rate limiting when enabled. Streamable HTTP uses MCP SDK behavior and differs from SSE. External wiki folders and mounted config.yaml paths must exist and be accessible to the server process. Vector search depends on optional vector/embedding dependencies and falls back where unavailable.