Tools · Memory API
Supermemory Explained
Supermemory is a managed memory API and MCP server for AI apps and agents — offering simple integration for storing, retrieving and sharing context across Claude, Cursor, Windsurf and custom agent stacks without building extraction pipelines from scratch.
Supermemory layers
Definition
What is Supermemory?
A managed context and memory infrastructure platform — REST API plus MCP server for persistent agent memory across sessions and tools.
Supermemory (supermemory.ai) provides five layers: user profiles, a memory graph, hybrid retrieval, format extractors (PDF, web, audio) and data connectors (Notion, Google Drive, Gmail). Core API endpoints include POST https://api.supermemory.ai/v3/add and POST https://api.supermemory.ai/v3/search (Supermemory docs, 2026).
The open-source repo (supermemoryai/supermemory, MIT license) includes a local mode via npx supermemory local for private deployments. Validate recall on your own eval set before production.
MCP
Supermemory MCP integration
MCP Server 4.0 gives Claude, Cursor, Windsurf and other MCP clients persistent memory via https://mcp.supermemory.ai/mcp.
Quick install (Supermemory docs):
npx -y install-mcp@latest https://mcp.supermemory.ai/mcp --client cursor --oauth=yesMCP tools include memory (save/forget), recall (search + profile) and context (inject user profile as system context). Auth via OAuth or API keys prefixed sm_. Project scoping via x-sm-project header.
Comparison
Supermemory vs Engram and alternatives
Supermemory leads on MCP/IDE integration; Engram leads on Weaviate-native production stacks.
| Tool | Strength | Best for |
|---|---|---|
| Engram | Weaviate-native memory layer | Weaviate stacks |
| Supermemory | MCP + IDE plugins + memory graph | Cursor/Claude MCP workflows, fast experiments |
| Zep | Temporal knowledge graph | Time-changing facts |
| Letta | Virtual-context paging | Long conversations |
Fit
When to use Supermemory
Fast experiments, MCP-first workflows and API-simple apps — skip when you need Weaviate-native ops or temporal graph semantics.
- Use Supermemory — Cursor/Claude MCP setup, personal cross-app memory, connector sync (Notion, Drive), prototype apps on free tier
- Use Engram — Weaviate-native production stack, scalable memory layer
- Use Zep — temporal invalidation, enterprise graph memory
- Use Letta — virtual-context paging for very long threads
FAQ
Frequently asked questions
Supermemory vs Mem0?
Mem0 has published LOCOMO J 66.9 (Chhikara et al., 2025) and broader framework integration. Supermemory leads on MCP/IDE plugins and connector sync. See Mem0 alternatives.
Does Supermemory support MCP?
Yes — MCP Server 4.0 at https://mcp.supermemory.ai/mcp. Tools: memory, recall, context. Works with Claude, Cursor, Windsurf (Supermemory docs, 2026).
Supermemory pricing?
Free tier at $0/mo for prototypes (Supermemory docs, 2026). Paid tiers for production scale — check supermemory.ai for current plans.
Is Supermemory open source?
The GitHub repo (supermemoryai/supermemory) is MIT-licensed. Managed cloud API and MCP are hosted services. Local mode: npx supermemory local.
Supermemory LOCOMO benchmarks?
No public LOCOMO or LongMemEval scores published as of July 2026. Run your own recall eval before production. Mem0 and Zep have published benchmark numbers.
Supermemory at production scale?
Offers SOC 2, HIPAA, GDPR compliance and self-hosting options (Supermemory docs). Validate latency and recall on your workload — hybrid retrieval claims sub-300ms in their docs.