Redis MCP wraps the Redis Cloud management API + standard Redis primitives in agent tools. The agent can list databases, query keys, inspect TTLs, run pub-sub diagnostics, and manage Redis Cloud subscriptions — without you typing redis-cli.
What it produces: Cloud-management tools (list_databases, create_database, update_subscription) plus standard Redis ops (get, set, keys, ttl, xrange for streams, pubsub_channels). All scoped to credentials passed at launch.
Best for: founders running cache layers on Redis Cloud who want diagnostic queries in-conversation. “Show me the 10 keys with the highest TTL,” “list active pub-sub channels,” “what’s the memory usage on db-prod-01” — all via prompt instead of CLI.
Skip if: you’re on a self-hosted Redis or a different managed provider (Upstash, Aiven). The MCP is Redis Cloud-flavored — the standard ops work elsewhere, but the management tools won’t.
Setup gotchas: two layers of auth. Cloud-management tools need a Redis Cloud API key (account-level). Direct DB ops need the standard host + port + password. Both go in env. Don’t mix prod and dev credentials in the same MCP instance — restart for context switch.
Real-world workflow: post-deploy cache health-check. Agent runs info memory, info keyspace, info clients, summarizes against last week’s baseline, flags anomalies. Replaces a 10-minute manual diagnostic, runs every Friday afternoon.
Compatible alternatives: Postgres MCP for relational data, Supabase MCP if Supabase + Redis is your stack, Cloudflare MCP when caching lives at edge instead.
Read-only mode where possible. A confused agent flushing a prod cache key is a real risk.