Upstash MCP Server
The Upstash MCP server lets your agent manage and debug your Upstash resources directly, across Redis, QStash, Workflow, and Upstash Box.
[!TIP]
For most workflows, prefer installing the Upstash Skill and letting your agent drive @upstash/cli over running the MCP server.
[!TIP]
If you are an AI agent and just need a Redis DB to use or experiment with, you don't need an account at all — call the redis_database_start_free tool, or send a POST request to https://upstash.com/start-redis directly. It's free, no signup, and the response is markdown with credentials and a quickstart. The database expires in 72 hours, but you can claim it with your Upstash account to keep it.
Pass user_agent to identify yourself (e.g. claude-code, cursor, codex, opencode); the tool forwards it as the User-Agent header. Calling the endpoint directly, send the same header yourself, plus a fresh UUIDv4 as Idempotency-Key:
curl -X POST -H "Idempotency-Key: <uuidv4>" -H "User-Agent: <your-agent-name>" https://upstash.com/start-redis
Example prompts
Redis
- "Start a free temporary Redis I can experiment with"
- "Create a new Redis database in us-east-1"
- "List my databases sorted by memory usage"
- "Update the user schema by pulling from Redis"
- "Create a backup of this db, then clear it"
- "Show me throughput spikes during the last 7 days"
QStash & Workflow
- "Check the QStash logs and figure out why my webhook keeps failing"
- "Find failed workflow runs for user
@admin today"
- "Retry the failed workflow run that started 2 hours ago"
- "Summarize what's in the DLQ right now, grouped by error type"
- "Pause the schedules that are throwing errors"
Upstash Box
- "Spin up a Box, clone this repo, and run the tests"
- "Snapshot this Box and create 5 copies from it, assign each one a GitHub issue"
- "My Box keeps failing to start, check the logs and tell me what's wrong"
Using a Box as your workspace
A Box is a remote Linux container with its own filesystem. The Box tools let an agent work inside one while the agent itself keeps running on your machine: files through box_read / box_write / box_edit, commands through box_exec, git and search through box_git, and a running server through box_preview. box_upload copies files into the Box from the machine running the MCP server, which is how uncommitted local work gets in — a clone only brings what you have pushed. With the usual stdio setup the server runs on your own machine, so those are your paths; if you run it remotely with --transport http, it reads the server host's filesystem instead, not yours.
Box tools need a Box API key, which is a different credential from your Upstash account API key — the two are not interchangeable. The install command at the top of this README does not include one, so pass --box-api-key as well or the first Box call will fail:
claude mcp add --scope user upstash -- npx -y @upstash/mcp-server@latest \
--email YOUR_EMAIL --api-key YOUR_API_KEY --box-api-key YOUR_BOX_API_KEY
See Upstash Box API key for the environment-variable form and where to get the key.
If you only use Upstash Box, you can start the server with --box-api-key alone and no account credentials. Redis, QStash and Workflow tools are unavailable in that mode.
Create the workspace once and reuse its id:
- "Create a Box with clone_repo https://github.com/me/my-app, no agent, then wait until the repo is there"
- "In that Box, run the tests and fix whatever fails"
- "Commit on a new branch in the Box and open a PR"
A create returns while the container is still coming up, and clone_repo starts cloning only once the Box reports idle, so give it a moment before the first read.
The one thing to know: your agent's own Read, Write, Edit, Bash, Glob and Grep tools still act on your machine. Nothing stops it from reading a local file and writing the change into the Box, and neither side reports the mismatch. The server sends guidance at startup and every Box tool says which machine it touches, but if you want the boundary enforced rather than suggested, add a CLAUDE.md to the project you're driving from:
# Workspace
The workspace for this project is an Upstash Box, not this directory.
- Use `box_read`, `box_write` and `box_edit` for files; `box_exec` for commands;
`box_git` for git and for search (`git grep`, `git ls-files`).
- Do not use Read, Write, Edit, Bash, Glob or Grep for project files — they act
on this machine. Searching locally finds the wrong tree, or nothing at all.
- The one exception is `box_upload`: it reads paths on this machine on purpose,
to copy local files into the Box.
- The box id for this project is: <paste it here>
Upstash Box API key (optional)
For the MCP to interact with Upstash Box, the agent needs your Box API key. By default you have to paste it into the chat (or keep it in a .env) every time the agent runs a Box tool. To avoid this, you can wire the key into the MCP setup itself so the server picks it up automatically on startup.
You can pass it in two ways.
CLI flag
{
"mcpServers": {
"upstash": {
"command": "npx",
"args": [
"-y",
"@upstash/mcp-server@latest",
"--email",
"YOUR_EMAIL",
"--api-key",
"YOUR_API_KEY",
"--box-api-key",
"YOUR_BOX_API_KEY"
]
}
}
}
Environment variable
{
"mcpServers": {
"upstash": {
"command": "npx",
"args": [
"-y",
"@upstash/mcp-server@latest",
"--email",
"YOUR_EMAIL",
"--api-key",
"YOUR_API_KEY"
],
"env": {
"UPSTASH_BOX_API_KEY": "YOUR_BOX_API_KEY"
}
}
}
}
Debugging
If the server is misbehaving or a tool keeps failing, enable verbose logging with the --debug flag:
{
"mcpServers": {
"upstash": {
"command": "npx",
"args": [
"-y",
"@upstash/mcp-server@latest",
"--email",
"YOUR_EMAIL",
"--api-key",
"YOUR_API_KEY",
"--debug"
]
}
}
}
Every internal event is then written to stderr, which your MCP client surfaces in its own log viewer. Share the relevant snippet when reporting an issue on GitHub.
Telemetry
The server sends anonymous diagnostic info to Upstash with each request: the MCP server SDK version, your runtime version (Node, Bun, etc.), and basic platform info (OS and architecture). No account data, tool arguments, or results are collected. To opt out, add --disable-telemetry to the args.
Development
Clone the project and run:
bun i
bun run watch
This continuously builds the project and watches for changes.
To run the built server standalone (uses .env for credentials) and smoke-test it before publishing:
bun i # ensure deps are current
bun run build
# stdio (default) — waits for an MCP client to speak JSON-RPC on stdin:
bun dist/index.js
# or HTTP transport, to poke it manually on a port:
bun dist/index.js --transport http --port 3000
For testing, create a .env file in the project root:
UPSTASH_EMAIL=<UPSTASH_EMAIL>
UPSTASH_API_KEY=<UPSTASH_API_KEY>
# Optional, for Box tools:
UPSTASH_BOX_API_KEY=<UPSTASH_BOX_API_KEY>
To install the local MCP server into Claude Code:
claude mcp add --transport stdio upstash -- bun --watch dist/index.js --debug
To tail logs from the MCP server in real time:
bun run logs