Upstash MCP. Manage Caches, Queues, and State in Redis from Chat.

Your AI client connects directly to Upstash Redis, letting your agent treat complex data structures—like caches, message queues, or user state—as native functions in the conversation. You don't gotta open a command line; your agent runs atomic commands like lrange, hset, and pipeline instantly.

Managing Basic Keys and Strings:
You can use get to pull out the string value tied to any specific Redis key, or you can use set to write or update that simple text value. When you set a key, you can even specify an expiration time, giving it a Time To Live (TTL), which makes sure it cleans itself up automatically when it hits zero seconds.

If you need to change a counter, you've got incr, which increases the number by one—starting at zero if the key is brand new—or decr to count it down. You can also check if a key exists using exists before trying to read it, and if you're done with a piece of data, del permanently wipes that key from the database.

Working with Structured Data (Hashes):
For things like user profiles or complex records where one key holds multiple related pieces of info, you use hashes. You can set or overwrite individual fields inside a hash using hset. If you need to read just one specific field from that structured record, you pull it out with hget, but if you want everything—all the fields and their values—you call hgetall.

This keeps all related data neatly bundled under a single key.

Building Queues and Ordered Lists:
When you're dealing with job queues or ordered sequences, lists are your best friend. You can add items to either end of the list: use lpush to stick values on the left side (the head), or use rpush to tack them onto the right side (the tail).

To pull data out for processing, you use lrange, which retrieves a defined section of elements from that list. You can always check how many items are sitting in the queue using llen.

Tracking Unique Membership (Sets):
If your data needs to guarantee uniqueness—like tracking user permissions or unique tags—you gotta use sets. To add members, you call sadd, and it automatically ignores any duplicates; that's its superpower. You check if a specific member is already in the set super fast with sismember (it’s constant time, so it doesn't matter how big your database gets).

If you need to grab all the unique members stored there, use smembers. And remember, if those tags or permissions are stale, you can remove them using srem.

Advanced Operations and Efficiency:
For communication, you send messages out using publish, broadcasting a message to everyone listening on that specific Redis channel. To handle batches—say, checking existence, setting a key, and getting all the members of a set—you wrap it up into a single network request with the pipeline tool.

This is huge for throughput because you don't gotta wait for each command to run one after the other.

Beyond these core tools, your agent can figure out what kind of data a key holds using key_type. It also lets you list keys based on patterns or prefixes with list_keys, though you should use that sparingly when working with huge databases. If you're dealing with complex transaction logic, remembering to check the current TTL with ttl helps you anticipate when data will disappear.

Your agent handles all this: it processes ordered lists for queues, manages unique membership in sets, and uses hashes to keep related metadata together. It’s a full-stack memory layer for your conversational AI client.