Mem0 MCP Server
OfficialArchived official Mem0 server for long-term agent memory: add, search, update, and delete memories via the Mem0 API.
Add to your client
Copy the config for your MCP client and paste it into its config file.
pip install mem0-mcp-serverPaste into ~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"mem0-mcp-server": {
"command": "uvx",
"args": [
"mem0-mcp-server"
],
"env": {
"MEM0_API_KEY": "<your-mem0-api-key>",
"MEM0_DEFAULT_USER_ID": "<your-user-id>"
}
}
}
}Requires `uv` (the Python package runner). Install it from https://docs.astral.sh/uv/ if `uvx` is not found.
Step-by-step guides: Add to Claude Desktop · Add to Cursor · Add to Windsurf
Before you start
- A Mem0 platform account and API key (m0-...) from app.mem0.ai
- uv/uvx (recommended) or pip with Python to run mem0-mcp-server
- Optional: MEM0_DEFAULT_USER_ID env var to scope memories to a user handle (defaults to 'mem0-mcp')
About Mem0 MCP Server
mem0-mcp-server is a thin, official bridge between MCP clients and the Mem0 Memory API. You run it locally with uvx mem0-mcp-server (stdio), it authenticates to the Mem0 platform with your MEM0_API_KEY, and memories are stored and searched in Mem0's cloud. MEM0_DEFAULT_USER_ID scopes reads and writes to a user handle, and an optional graph-memory flag (MEM0_ENABLE_GRAPH_DEFAULT) turns on Mem0's graph features per request.
The tool surface is complete rather than minimal: add_memory accepts text or full conversation history, search_memories does semantic search with filters, get_memories paginates with structured filters, and update/delete operate per-memory by ID. Entity-level tools (list_entities, delete_entities) manage the users/agents/apps/runs namespaces that Mem0 organizes memories under. Docker and Smithery deployments were also supported, including an HTTP endpoint mode.
The headline caveat: the project is archived. Mem0's own banner recommends its cloud-hosted MCP server (Streamable HTTP at https://mcp.mem0.ai/mcp, added via npx mcp-add) as the successor for platform users. Within the same ecosystem, OpenMemory — shipped inside the mem0ai/mem0 monorepo — is the local-first alternative: same 'memory layer for MCP clients' idea, but the data stays on your machine with a dashboard, instead of round-tripping to the Mem0 cloud like this server does.
If you are picking today: hosted Mem0 MCP for the platform-managed path, OpenMemory (or Mem0's self-hosted server) for local control. mem0-mcp remains useful mainly for existing setups that depend on its exact tool names and its uvx/stdio shape.
Tools & capabilities (9)
add_memorySave text or conversation history (or explicit message objects) for a user/agent.
search_memoriesSemantic search across existing memories with filters and limit.
get_memoriesList memories with structured filters and pagination.
get_memoryRetrieve one memory by its memory_id.
update_memoryOverwrite a memory's text once the user confirms the memory_id.
delete_memoryDelete a single memory by memory_id.
delete_all_memoriesBulk delete all memories in the confirmed scope (user/agent/app/run).
delete_entitiesDelete a user/agent/app/run entity and its memories.
list_entitiesEnumerate users/agents/apps/runs stored in Mem0.
What this server can do
Mem0 MCP Server provides tools for these capabilities — tap one to see every MCP server that does the same:
When to use it
- Give Claude Desktop or Cursor persistent recall of user preferences and project facts across sessions
- Store structured research or trial parameters mid-conversation and retrieve them semantically later
- Manage memory hygiene: list, update, or bulk-delete memories per user/agent/app scope
- Keep an existing Mem0-based agent stack running unchanged despite the archive (with pinned versions)
Quick setup
- 1Get a Mem0 API key from the Mem0 platform dashboard
- 2pip install mem0-mcp-server (or uv pip install mem0-mcp-server)
- 3Add the uvx config with MEM0_API_KEY and MEM0_DEFAULT_USER_ID to your MCP client config
- 4Restart the client and test with something like 'remember that I prefer pytest over unittest'
- 5New project? Consider the hosted Mem0 MCP (https://mcp.mem0.ai/mcp) or OpenMemory instead — this repo gets no further updates
Security notes
This repo is archived (read-only since March 2026): no maintenance, no security patches — Mem0 now points users at its hosted MCP endpoint instead. The server sends memory content to the Mem0 cloud API under your MEM0_API_KEY, and delete_all_memories / delete_entities are destructive bulk operations, so scope the key carefully.
Mem0 MCP Server FAQ
Is mem0-mcp still safe to use?
It still works, but the repo has been a public archive since March 2026: no bug fixes, no security patches, no releases. For anything new, Mem0 recommends its hosted MCP server at https://mcp.mem0.ai/mcp; for local-first memory, OpenMemory in the mem0 monorepo is the sibling project.
What replaced mem0-mcp — and where does OpenMemory fit?
Two paths. The direct successor is Mem0's cloud-hosted MCP endpoint (https://mcp.mem0.ai/mcp), which needs no local process. OpenMemory is the local alternative from the same company: it runs the whole memory stack on your machine (Docker + dashboard), whereas mem0-mcp was a local shim that stored everything in the Mem0 cloud.
Where do my memories actually live?
In Mem0's cloud platform, not on your machine. The server is a stdio bridge that calls the Mem0 API with your MEM0_API_KEY, and responses are raw JSON from that API — so data residency and retention follow your Mem0 account, and revoking the key cuts off access.
Alternatives to Mem0 MCP Server
Compare all alternatives →
Official MCP server providing persistent, file-backed knowledge-graph memory across sessions.
Structured step-by-step reasoning tool for breaking problems into revisable thought sequences.
Mem0's local-first memory layer: a Dockerized MCP server plus dashboard that keeps agent memories on your machine.
Compare Mem0 MCP Server with: