Supermemory MCP Server
Give your AI agents persistent memory — store, recall, and connect facts across every conversation.
Updated
What is Supermemory MCP?
The Supermemory MCP server is a Model Context Protocol server that gives AI assistants persistent memory — letting them store, retrieve, and connect facts across conversations and applications.
Supermemory acts as a universal memory layer for LLMs, solving the problem of agents forgetting context between sessions. It can save new memories, run semantic search to recall relevant ones, and build up a user profile from what it has stored. The service is vendor-hosted on Cloudflare's platform, so there is nothing to deploy yourself — you connect to the remote endpoint at https://mcp.supermemory.ai/mcp.
Authentication uses OAuth by default: your MCP client discovers the authorization server automatically and prompts you to sign in with a Supermemory account. Programmatic access via an API key (prefixed sm_, from app.supermemory.ai) is also supported.
Tools the Supermemory MCP server exposes
| Tool | What it does |
|---|---|
| memory | Save (or forget) information to Supermemory; ingestion into the search index is eventually consistent. |
| recall | Run semantic search over stored memories and optionally retrieve the user profile. |
Connect the Supermemory MCP server
Claude Code
- 1
Add the remote server
Run: claude mcp add --transport http supermemory https://mcp.supermemory.ai/mcp
- 2
Authenticate via OAuth
On first use, Claude Code opens the Supermemory OAuth flow in your browser — sign in to authorize access.
- 3
Verify
Run /mcp in Claude Code to confirm the supermemory server is connected and its memory/recall tools are listed.
Cursor
- 1
Run the installer
npx install-mcp https://mcp.supermemory.ai/mcp --client cursor --oauth yes
- 2
Complete OAuth
The CLI launches the Supermemory OAuth flow; sign in to authorize. You can also add the server manually in ~/.cursor/mcp.json.
- 3
Confirm in Cursor
Open Cursor Settings → MCP and check that supermemory shows as connected.
Sample use cases
A scheduled research agent recalls what it already gathered before each run
Avoids duplicate work and builds on prior findings
A support or sales agent stores key facts about a contact mid-conversation
Future sessions remember the relationship without re-asking
Cross-tool memory shared between Claude, Cursor, and other LLM clients
One memory layer follows the user across every assistant
Security & permissions
Supermemory MCP uses OAuth by default — your MCP client discovers the authorization server and prompts you to sign in with a Supermemory account, so the agent acts within your account's memory scope (memories can be organized per project). An API key (sm_…, from app.supermemory.ai) can be passed in the Authorization header for programmatic access instead. The agent can read and write your stored memories and the auto-generated profile, so treat the connected account and key as sensitive.
Using Supermemory MCP with Gamut
In a Gamut workflow, an agent can use Supermemory as a connected tool to give automated, event- or schedule-triggered runs durable context. Gamut manages the OAuth connection, so a scheduled or webhook-triggered agent can call recall to pull relevant prior facts before acting and memory to persist new outcomes — letting recurring jobs build on what earlier runs learned instead of starting cold each time.
Frequently asked questions
What is the Supermemory MCP server?
It is a Model Context Protocol server that gives AI assistants persistent memory. It exposes a memory tool to store facts and a recall tool to semantically search them, so agents retain context across conversations and apps. It is hosted by Supermemory at https://mcp.supermemory.ai/mcp.
Is the Supermemory MCP server free?
Yes — Supermemory's free tier ($0/month, roughly $5 of monthly usage) includes the MCP server. Heavier or production use moves to usage-based paid plans (Pro $19/mo, Max $100/mo, Scale $399/mo) and custom Enterprise.
How do I connect Supermemory MCP to Claude or Cursor?
For Claude Code, run claude mcp add --transport http supermemory https://mcp.supermemory.ai/mcp and complete the OAuth sign-in. For Cursor, run npx install-mcp https://mcp.supermemory.ai/mcp --client cursor --oauth yes (or add the URL in ~/.cursor/mcp.json).
Is Supermemory MCP hosted or self-hosted?
The MCP server is vendor-hosted on Supermemory's Cloudflare-based infrastructure — you connect to the remote endpoint with OAuth, nothing to deploy. Self-hosted and air-gapped deployments are available on the Enterprise plan.