Redis MCP Server: Tools, Setup and Security

The Redis MCP server gives AI apps a natural-language way to store, read and search data in Redis. It covers strings, hashes, lists, sets, sorted sets, streams, pub/sub, JSON documents and vector indexes. Redis publishes it as an open-source Python package that you run locally. Redis's MCP docs page calls it "a general-purpose implementation that lets agents read, write, and query data in Redis."

Checked against: Redis MCP Server README on 28 October 2026. eSEOspace did not install or test this server.

At a glance

What it can do

Redis describes its tools as groups rather than listing every tool name:

• string: set and get strings with expiration, for config values, sessions or cached responses.
• hash: field-value pairs in one key. Hashes can hold vector embeddings.
• list: append and pop items, for queues or recent-activity lists.
• set: add, remove and list unique members.
• sorted set: score-ordered data for leaderboards or priority queues.
• pub/sub: publish messages, create channel or pattern subscriptions, and read queued messages.
• streams: add, read and delete entries, manage consumer groups and acknowledge entries.
• JSON: store, read and change JSON documents by path.
• query engine: manage vector indexes and run vector search.
• docs: search Redis documentation in natural language.
• server management: get information about the database.

Example prompts in the README include "Cache this item" and "Index and search this vector."

How to connect it

Redis recommends the PyPI package, run with uvx. The README gives this client config:

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

The --url follows the standard redis:// and rediss:// schemes. For an encrypted connection, Redis shows:

rediss://user:secret@hostname:port?ssl_cert_reqs=required&ssl_ca_certs=path_to_the_certificate

You can also run a tagged release straight from GitHub. Redis says to prefer a tagged release, because the main branch "may contain breaking changes." The README includes setup notes for Claude Desktop, VS Code with GitHub Copilot, Augment and the OpenAI Agents SDK.

Security notes

The Redis MCP server can write and delete data. It has no built-in read-only flag. The README's answer is Redis's own access control. It shows how to create a read-only user:

ACL SETUSER readonlyuser on >mypassword ~* +@read -@write

Connect the MCP server as that user, and Redis itself will refuse writes. You can go further and limit the key patterns the user can see. That is least privilege at the database level.

Other points to plan for:

• Credentials. The connection URL carries a username and password. Keep it in your client config or environment, not in prompts. See secrets and credential leakage in MCP.
• Certificate checks. Redis shows ssl_cert_reqs=none for unverified connections. Use required with a CA certificate in any real setup.
• Local process risk. As a local stdio server, it runs with your user's rights. Read about sandboxing local MCP servers.
• Pub/sub and streams. These can send messages to other systems that read from Redis. Treat them as write tools.

More guidance is in the MCP security hub.

Alternatives

• MongoDB MCP Server: document database tools with a --readOnly flag.
• Pinecone MCP Server: a vector database server for coding assistants.
• Elastic MCP Server: search and analytics through Elastic Agent Builder.

More profiles are in the MCP server directory.

Next step

Create a dedicated ACL user for the AI before you connect, and decide which key patterns it may touch. If you are building agents that use Redis for memory or caching, our MCP development team can help design that.

Frequently asked questions