Redis MCP server: connect your cache to Claude & Cursor (2026)

A Redis MCP server gives an AI tools to read and write your Redis data — strings, hashes, lists, sets, streams and even vector search — over natural language. Redis maintains an official server; here's the clean setup.

The pick: Redis's official MCP server

The official redis/mcp-redis server takes a standard Redis connection URL and exposes typed tools for the core data types plus search. It runs locally via uvx (Python) and points at any reachable Redis.

Step 1 — point it at your instance

The simplest form passes a connection URL:

uvx --from redis-mcp-server@latest redis-mcp-server --url redis://user:pass@localhost:6379/0

In a client config (Claude Desktop / Cursor):

{
  "mcpServers": {
    "redis": {
      "command": "uvx",
      "args": ["--from", "redis-mcp-server@latest", "redis-mcp-server", "--url", "redis://localhost:6379/0"]
    }
  }
}

You can also configure via env vars — REDIS_HOST, REDIS_PORT, REDIS_PWD, REDIS_SSL, REDIS_CLUSTER_MODE — instead of a URL. Restart the client; see MCP config file location.

Step 2 — verify

Ask: "How many keys match session:* and what's the TTL on the newest one?" A real count back confirms it works.

Common problems

uvx not found — install uv (the Python runner) first; it's the recommended way to run the server.
Connection refused — check host/port and that the URL credentials are right; for TLS set REDIS_SSL=true.
Cluster keys missing — enable REDIS_CLUSTER_MODE when pointing at a clustered deployment.

Security

A write-capable Redis tool can flush or overwrite data, so scope the ACL user the MCP connects as — read-only where you can, and never the default superuser in production. Keep connection strings out of shared configs. See MCP security best practices.

Going further

Redis works well as agent memory and a cache — compare it in best MCP server for a database and see the Redis agent profile. Browse the databases category or the data-analyst loadout.