cursor-mem0 0.1.0

AI agent memory layer (mem0-compatible) powered by Cursor SDK

cursor-mem

AI agent memory layer modeled on mem0, with one key change: LLM inference uses Cursor SDK and CURSOR_API_KEY instead of OpenAI / Anthropic / other third-party LLM API keys.

Everything else follows mem0’s pipeline (v3 additive extraction, hybrid retrieval, entity linking, SQLite history, Qdrant vector store).

Architecture

Component	Provider	API key
Memory extraction & reasoning	cursor-sdk Agent.prompt	CURSOR_API_KEY
Embeddings	fastembed (local ONNX)	None
Vector store	Qdrant (embedded, on disk)	None
History	SQLite	None

flowchart LR
  subgraph input [Conversation]
    M[messages]
  end
  subgraph cursor_mem [cursor-mem]
    S[search: embed + BM25 + entities]
    L[Cursor SDK LLM extract facts]
    V[(Qdrant)]
    H[(SQLite history)]
  end
  M --> S
  S --> L
  L --> V
  L --> H

Install

From PyPI (recommended):

pip install cursor-mem0

# MCP server for Cursor agents
pip install "cursor-mem0[mcp]"

# Optional: BM25 + entity extraction (same as mem0[nlp])
pip install "cursor-mem0[nlp]"
python -m spacy download en_core_web_sm

Note: PyPI package name is cursor-mem0 (import: from cursor_mem import Memory). The name cursor-mem on PyPI is a different project (IDE session hooks).

From source:

pip install -e .

# Optional extras
pip install -e ".[mcp]"
pip install -e ".[nlp]"
python -m spacy download en_core_web_sm

Setup

1. Create an API key: Cursor Dashboard → Integrations
2. Configure locally (recommended):

cp .env.example .env
# Edit .env and set CURSOR_API_KEY=...

Or export in the shell:

export CURSOR_API_KEY="cursor_..."

Data is stored under ~/.cursor-mem/ (override with CURSOR_MEM_DIR).

MCP in Cursor (project)

This repo includes a stdio MCP server so Cursor agents can call add_memory, search_memories, and list_memories.

1. Install with MCP extras

pip install "cursor-mem0[mcp]"
# or from source: pip install -e ".[mcp]"

2. Project config (already in repo)

File: .cursor/mcp.json

{
  "mcpServers": {
    "cursor-mem": {
      "command": "python3",
      "args": ["-m", "cursor_mem.mcp_server"],
      "cwd": "${workspaceFolder}",
      "env": {
        "CURSOR_API_KEY": "${env:CURSOR_API_KEY}",
        "CURSOR_MEM_USER_ID": "${env:CURSOR_MEM_USER_ID}"
      }
    }
  }
}

The server loads CURSOR_API_KEY from .env in the project root when Cursor starts it (cwd = workspace).

3. Enable in Cursor

1. Open this folder as the workspace (cursor-mem).
2. Cursor Settings → MCP (or run /mcp in chat) and confirm cursor-mem appears and is enabled.
3. If tools do not show up, reload the window and check MCP logs for python3 / missing mcp package.

4. Optional: user-global MCP

To use memory in every project, add the same block to ~/.cursor/mcp.json and set cwd to this repo’s absolute path, or export CURSOR_API_KEY globally.

MCP tools

Tool	Purpose
add_memory	Store facts from conversation (infer=true uses mem0 extraction via Cursor SDK)
search_memories	Semantic + hybrid search
list_memories	List memories for a user_id
get_memory	Fetch one memory by id

Quickstart

Same API as mem0’s Memory class:

from cursor_mem import Memory

memory = Memory()

# Add memories from a conversation
memory.add(
    "I prefer dark mode and vim keybindings",
    user_id="alice",
)

# Search
results = memory.search(
    "What are Alice's editor preferences?",
    filters={"user_id": "alice"},
    top_k=3,
)
for item in results["results"]:
    print(item["memory"], item.get("score"))

Chat example

export CURSOR_API_KEY="cursor_..."
python examples/chat_with_memory.py

Configuration

Defaults (Memory())

• LLM: provider: cursor, model composer-2.5, local.cwd = current directory
• Embedder: fastembed + thenlper/gte-large (1024 dims)
• Vector store: local Qdrant at ~/.cursor-mem/qdrant

Custom config (mem0-style dict)

from cursor_mem import Memory

memory = Memory.from_config({
    "llm": {
        "provider": "cursor",
        "config": {
            "api_key": "cursor_...",
            "model": "composer-2.5",
            "cwd": "/path/to/project",
        },
    },
    "embedder": {
        "provider": "fastembed",
        "config": {"model": "thenlper/gte-large"},
    },
    "vector_store": {
        "provider": "qdrant",
        "config": {
            "path": "/tmp/my-qdrant",
            "collection_name": "my_memories",
            "embedding_model_dims": 1024,
        },
    },
})

Environment variables

Variable	Purpose
CURSOR_API_KEY	Cursor SDK authentication (required for LLM)
CURSOR_MEM_DIR	Storage root (default ~/.cursor-mem)
CURSOR_MEM_USER_ID	Default user id in the example script

How it maps to mem0

• Unchanged: add, search, get, get_all, update, delete, prompts, scoring, entity store, history DB.
• Swapped: mem0.llms.openai.OpenAILLM → cursor_mem.llms.cursor.CursorLLM (registered as provider cursor).
• Default embedder: openai → fastembed so you do not need OPENAI_API_KEY for vectors.

Why fastembed for embeddings?

Cursor SDK exposes agent/LLM APIs, not a dedicated embeddings endpoint. mem0 still needs dense vectors for semantic search; fastembed runs locally with no cloud API key, which matches the goal of avoiding third-party keys while keeping mem0’s retrieval logic intact.

License

Apache-2.0 (aligned with mem0 OSS).