Mem0 MCP server
Mem0's hosted MCP server: add, search, update, and delete long-term memories in your Mem0 Platform account, scoped by user, agent, app, or run.
Mem0's MCP server is a hosted endpoint at https://mcp.mem0.ai/mcp that fronts the Mem0 Platform API. The agent sends text or a conversation to add_memory, Mem0 extracts discrete memories from it with an LLM, and later calls to search_memories run semantic search with filters over what was stored. Every memory belongs to an entity: a user_id, agent_id, app_id, or run_id inside your Mem0 project, and list_entities and delete_entities operate on those. Writes are processed asynchronously, which is why the server also exposes list_events and get_event_status. Nothing runs on your machine; memories live in your Mem0 account.
The hosted server is the current path. The earlier self-run package, mem0-mcp-server (mem0ai/mem0-mcp), is archived, and OpenMemory, Mem0's local MCP server with its own dashboard, was sunset and removed from the mem0ai/mem0 monorepo in July 2026; a read-only copy lives in mem0ai/openmemory. The Mem0 library itself is Apache-2.0 and can be self-hosted through the repo's server/ directory (FastAPI, PostgreSQL with pgvector, Neo4j), but the hosted MCP server's own source is not public. Memory is written only when the agent decides to call a tool, and it is partitioned by the IDs the agent passes rather than shared across a team by default.
Quick install
Copy-paste configs are provided for all 8 supported clients. Pick your client below.
Add to ~/.claude.json
{
"mcpServers": {
"mem0": {
"type": "http",
"url": "https://mcp.mem0.ai/mcp"
}
}
}claude mcp add --transport http mem0 https://mcp.mem0.ai/mcpHeads up
- First tool call opens a browser to authorize.
Available tools
| Tool | Description |
|---|---|
| add_memory | Save text or conversation history as memories for a user, agent, app, or run. |
| search_memories | Semantic search across stored memories, with filters and a result limit. |
| get_memories | List memories with structured filters and pagination. |
| get_memory | Retrieve one memory by its memory_id. |
| update_memory | Overwrite a memory's text or metadata once its memory_id is confirmed. |
| delete_memory | Delete a single memory by memory_id. |
| delete_all_memories | Bulk delete every memory in the confirmed user, agent, app, or run scope. |
| delete_entities | Delete a user, agent, app, or run entity together with its memories. |
| list_entities | Enumerate the users, agents, apps, and runs that hold memories in Mem0. |
| list_events | List memory operation events with filters and pagination. |
| get_event_status | Check the status of an asynchronous memory operation by event_id. |
Required configuration
- MEM0_API_KEYOptional
Mem0 Platform API key, sent as a bearer token by clients that cannot complete the browser sign-in (CI, Codex config.toml).
What you can do with it
Carry user preferences across clients
One Mem0 account backs Claude, Cursor, Codex, and VS Code, so a preference saved with add_memory in one client comes back from search_memories in another.
Per-user memory in an agent product
An agent you build passes a user_id on every call, which keeps each customer's memories in their own partition and lets you wipe one with delete_entities.
Audit what the agent wrote
Because writes are asynchronous, list_events and get_event_status show which add or update operations finished, and the Mem0 dashboard shows the resulting memories.
FAQ
- Is it free?
- Mem0 Platform has a free tier; paid plans are usage-based or custom. The MCP server itself costs nothing extra, but every memory it stores counts against your Mem0 account. The Mem0 library is open source under Apache-2.0.
- Does it support remote or OAuth?
- Yes. It is remote only, served over HTTPS at https://mcp.mem0.ai/mcp. Clients with browser sign-in complete an OAuth flow on first use; others send a Mem0 API key as a bearer token. Without either, the server returns 401.
- How does it compare to a team memory service like Glen?
- Mem0 stores what the agent chooses to save, partitioned by the user, agent, app, or run IDs it passes. Glen records every agent session and connected tool automatically and makes the result searchable across the whole org, with access checked per observation at recall.