DeepMem MCP
A long-term memory server for agents with a Mem0-compatible API, hybrid search, its own MCP server and Docker deployment.
- 114 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Conversation text goes to the chosen language model and to the deepmem.dev cloud if it is connected.
- The self-run server in open mode checks no keys and separates users only by `user_id`.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/deepmemDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Conversation text goes to the chosen language model and to the deepmem.dev cloud if it is connected.
- The self-run server in open mode checks no keys and separates users only by `user_id`.
This is third-party code. Review the repository files before installing.
What it does
DeepMem turns conversations into memory that outlives sessions. A FastAPI HTTP service sits in front of the Qdrant vector store: a language model extracts facts from messages, and search combines vector similarity, BM25 keyword match, entity boost and time decay. Writes collect behind a silence window and are processed in batches, while repeated requests answer from a semantic cache. The MCP server gives Claude Desktop, Cursor and other clients the tools `deepmem_write`, `deepmem_search` and `deepmem_delete`. The API mirrors Mem0 and the deepmem-client package copies MemoryClient, so leaving Mem0 comes down to one import line. Run the server yourself or use the deepmem.dev cloud on the same engine.
Who it is for. Agent developers who want cross-session memory on their own server without rewriting code written for Mem0.
Good fit when
- The agent keeps asking who the user is and what they work on.
- A project already calls Mem0 and should move to a self-run server.
- Decisions and root causes found earlier must be recalled in later Claude Code or Cursor sessions.
- Preferences of different end users are stored separately by `user_id`.
Not a fit when
- You need a knowledge graph with temporal links: DeepMem has none.
- Memory must hold large code fragments: long fenced blocks are filtered out.
- Requests with personal data cannot go to an external language model provider.
Example request
Remember that the project uses Postgres 16 and Docker deployment, and remind me of that at the start of the next session.MCP
- Transport
- http, stdio
- Authentication
- key is optional
Environment variables
`DEEPMEMORY_BASE_URL`- Server address for the local MCP, for example http://localhost:8000.
`DEEPMEMORY_API_KEY`secret- API key, needed for the cloud and empty for the open local mode.
Limitations
The author calls the approach hybrid retrieval rather than a knowledge graph and points to Zep for temporal graphs. Fact extraction needs your own keys: a language model (OpenAI, Anthropic or compatible) and embeddings (BGE-M3, Google or compatible). The self-hosted server runs in open mode without API keys and isolates users only by `user_id` in the request body, so it must not be exposed publicly without protection. The claimed twofold speedup and tenfold cost cut were measured by the author, with the method in benchmarks/README.md. The BGE-M3 embedder wants a GPU or patience on the CPU.
How to disable. Remove the deepmem entry from `.mcp.json` or the client settings and stop the containers with `docker compose down`. Delete Qdrant data and the cloud key separately.
FAQ
How do I connect DeepMem to Claude Code?
Run `claude mcp add --transport http deepmem https://deepmem.dev/mcp --header "Authorization: Bearer dm_live_..."` with your key. For a self-run server swap in your own address.
How does the self-run server differ from the cloud?
Engine and API are the same. The cloud needs a deepmem.dev key, while your own server starts without registration and keeps data with you.
How do I migrate from Mem0?
Install `pip install deepmem-client` and swap the `mem0` import for `deepmem`. MemoryClient method signatures are kept.
Which tools does the agent see?
The cloud exposes `deepmem_write` and `deepmem_search`, and the self-run MCP server adds `deepmem_delete`.
What does the memory store?
Preferences, facts and context. Large code blocks do not enter memory.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server for long-term agent memory: compresses dialogues into atomic facts, resolves pronouns and anchors relative time to absolute dates
Long-term memory for AI agents: a skill and the mindmemos command store facts from conversations and recall them in later sessions
A self-hosted MCP memory server that keeps preferences and decisions across conversations in Claude Code, Cursor and other clients
Long-term memory for agents: a remote MCP server, plugins for Claude Code, Cursor and Codex, and an API for your own apps
Membase keeps one shared memory for AI agents that Claude Code, Cursor and Codex read through MCP, an SDK or the membase command
Long-term agent memory with per-character profiles and video understanding: an MCP server and a drop-in replacement for Mem0
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| DeepMem (this card) | no | key is optional | 114 | |
| Mem0 | no | 68,887 | the original Mem0 memory layer with plugins and SDKs; DeepMem mirrors its API and adds its own MCP server and registration-free startup | |
| ai-memory | no | API key | 15,717 | keeps session experience in a Markdown wiki; DeepMem holds facts in a vector store with hybrid search |
README in short
One engine runs in the cloud and on your own server under one API. The author states it is hybrid retrieval without a knowledge graph and publishes scripts to repeat the benchmarks.