Graph Memory MCP

Self-hosted MCP server for graph-based long-term memory (FalkorDB or ArcadeDB). Agents decide what to write. The server deduplicates on insert, marks outdated facts, surfaces contradictions, and gives each owner_id its own graph or database. Embeddings run locally. The server makes no LLM calls.

Install

uvx graph-memory-mcp --host 127.0.0.1 --port 8000

pip install graph-memory-mcp
graph-memory-mcp --host 127.0.0.1 --port 8000

Streamable HTTP: http://127.0.0.1:8000/mcp. The first start downloads the default embedding weights (about 1.1 GB).

Database

ArcadeDB is the default. The image password is playwithdata. The process default for ARCADE_PASSWORD is empty, so set it.

docker run -d --name graph-memory-arcadedb -p 2480:2480 \
  -e JAVA_OPTS="-Darcadedb.server.rootPassword=playwithdata" \
  arcadedata/arcadedb:latest

ARCADE_PASSWORD=playwithdata graph-memory-mcp --host 127.0.0.1 --port 8000

FalkorDB:

docker run -d --name graph-memory-falkordb -p 6379:6379 \
  -e REDIS_ARGS="--requirepass falkordb123" \
  falkordb/falkordb

GRAPH_BACKEND=falkordb FALKORDB_PASSWORD=falkordb123 graph-memory-mcp --host 127.0.0.1 --port 8000

The same services are in docker-compose.yml.

MCP client

{
  "mcpServers": {
    "graph-memory-mcp": {
      "url": "http://127.0.0.1:8000/mcp"
    }
  }
}

Documentation

• README
• Configuration
• API contract
• Agent policies
• FAQ
• Comparison

MIT License.