Model Context Protocol (MCP)

Suprflo ships an MCP server that exposes your memory layer as MCP tools. Any MCP client (Claude Code, Claude Desktop, Cursor, Windsurf, Continue, Cline, or a custom client) can connect to it and read, write, and search memories, plus query the knowledge graph, without writing any HTTP code.

The server runs as a separate process. It speaks stdio JSON-RPC to the client and calls the Suprflo REST API over HTTP under the hood, so everything it does maps to the endpoints in rest-api.md.

Tools

The server exposes nine tools: five for memory operations and four for the knowledge graph.

Memory tools

Tool	Required args	Optional args	Purpose
add_memory	messages (array of strings), user_id	app_id	Extract and store memories from one or more messages. Maps to POST /api/memories.
search_memory	query, user_id	app_id, top_k (default 10), memory_types (semantic, episodic, procedural)	Semantic search over stored memories. Maps to POST /api/memories/search.
get_memory	memory_id	—	Retrieve a single memory by its id. Maps to GET /api/memories/{id}.
list_memories	user_id	app_id, limit (default 50)	List all memories for a user. Maps to GET /api/memories.
delete_memory	memory_id	—	Delete a specific memory. Maps to DELETE /api/memories/{id}.

Knowledge graph tools

Tool	Required args	Optional args	Purpose
query_graph	entity_name	max_depth (default 2)	Traverse the memory knowledge graph from an entity and return its neighborhood (connected entities and relationships).
get_communities	—	—	List all communities in the graph with entity counts.
get_god_nodes	—	top_k (default 5)	Return the most connected entities (god nodes) in the graph.
shortest_path	source, target	—	Find the shortest path between two entities in the graph.

Environment variables

The server reads its configuration from the environment at startup:

Variable	Default	Description
MEMORYG_API_URL	http://localhost:8000	Base URL of the Suprflo API. Set this to https://api.suprflo.com for the hosted service; the localhost default is only for a self-hosted backend.
MEMORYG_API_KEY	(empty)	API key sent as Authorization: Bearer <key>.
MEMORYG_ORG_ID	(empty)	Optional organization id, sent as X-Org-Id when set.
MEMORYG_PROJECT_ID	(empty)	Optional project id, sent as X-Project-Id when set.

For the hosted API, MEMORYG_API_KEY already carries your organization and project scope, so MEMORYG_ORG_ID and MEMORYG_PROJECT_ID are not needed.

Configure the server

Add the server to your client's MCP config. The shape is the same everywhere: command is python, args points at mcp_server.py, and env supplies the API URL and key.

{
  "mcpServers": {
    "memoryg": {
      "command": "python",
      "args": ["mcp_server.py"],
      "cwd": "/path/to/mem0-app/backend",
      "env": {
        "MEMORYG_API_URL": "http://localhost:8000",
        "MEMORYG_API_KEY": "your-api-key-here"
      }
    }
  }
}

The mcp and httpx packages must be available to the Python you point at:

pip install mcp httpx

Claude Code example

Claude Code auto-discovers MCP servers from a .mcp.json file in the project root. Point it at the hosted API and drop in your key:

{
  "mcpServers": {
    "suprflo": {
      "command": "python",
      "args": ["/path/to/mem0-app/backend/mcp_server.py"],
      "env": {
        "MEMORYG_API_URL": "https://api.suprflo.com",
        "MEMORYG_API_KEY": "<YOUR_API_KEY>"
      }
    }
  }
}

Restart Claude Code, then run /mcp to confirm the suprflo server is connected and lists nine tools. From there just talk naturally and Claude Code routes to the right tool:

> search alice's memories for anything about databases
> add a memory: Alice prefers dark mode
> what are the most connected entities in the graph?
> find the path between FastAPI and PostgreSQL