> ## Documentation index
> Fetch the complete documentation index at: https://upstash.com/docs/llms.txt
> Search it with GET https://upstash.com/docs/search?q=<query>.
> Use these to discover all available pages before exploring further.

# AI Tools

> Connect your AI coding agent to Upstash.

Upstash ships Skills, a CLI, an MCP server, and llms.txt so your agent can build on and manage Redis, QStash, Workflow, Vector, Search, and Box. Set them all up with one command, or pick your agent for step-by-step instructions.

# Set up with one command

Run this in your terminal:

```bash
npx upstash
```

It detects Claude Code, Codex, Cursor, Gemini CLI, VS Code, GitHub Copilot CLI, and OpenCode, and lets you pick which to set up. Each one gets the Upstash [MCP server](/agent-resources/mcp) and [skills](/agent-resources/skills): as the Upstash plugin where the agent supports plugins, otherwise written into its config. The MCP server signs in with OAuth on first use. See the [CLI page](/agent-resources/cli#connect-your-ai-agent) for flags.

# Pick your agent

<AgentGrid />

Don't see your agent? See the [full list of agents](/agent-resources/clients).

# Resources

<CardGroup cols={2}>
  <Card title="Skills" icon="sparkles" href="/agent-resources/skills">
    Docs and CLI reference for every Upstash SDK, loaded on demand.
  </Card>
  <Card title="MCP Server" icon="plug" href="/agent-resources/mcp">
    Live tools to query, debug, and manage your Upstash account.
  </Card>
  <Card title="CLI" icon="terminal" href="/agent-resources/cli">
    Agent-friendly terminal for managing resources. JSON output, dry runs.
  </Card>
  <Card title="llms.txt" icon="file-lines" href="/llms.txt">
    The docs as plain text for tools without skill or MCP support.
  </Card>
</CardGroup>

# Set up by hand

Install the Upstash skill. It includes the `upstash` CLI reference, so this one step is enough for most workflows:

```bash
npx skills add upstash/skills
```

To let your agent read and change your account directly, add the MCP server. The **remote** server needs nothing installed — connect over HTTP and approve the OAuth consent on first use (shown here for Claude Code):

```bash
claude mcp add --scope user --transport http upstash https://mcp.upstash.com/mcp
```

Prefer a headless setup, or need [Upstash Box](/box/overall/quickstart) tools? Run the local server with a [Developer API key](/devops/developer-api/introduction#create-an-api-key):

```bash
npx -y @upstash/mcp-server@latest --email YOUR_EMAIL --api-key YOUR_API_KEY
```

See the [MCP server page](/agent-resources/mcp) for remote vs. local, authentication, and feature options.

<Note>
  Only working against a single Redis database? The lightweight [Redis MCP](/redis/sdks/mcp) needs just that database's REST URL and token, not an account API key.
</Note>

# Building AI into your app?

The resources above are for agents that help you build. If you're building agents or AI features into your own product, start here:

<CardGroup cols={2}>
  <Card title="Agent memory & chat history" icon="brain" href="/redis/sdks/agentkit/ai-sdk">
    Long-term memory, RAG, and chat history for the Vercel AI SDK and Eve, backed by Redis.
  </Card>
  <Card title="Vector" icon="diagram-project" href="/vector/overall/getstarted">
    Embeddings and semantic search for RAG pipelines, with LangChain, LlamaIndex, and AI SDK integrations.
  </Card>
  <Card title="Workflow Agents" icon="robot" href="/workflow/agents/overview">
    Durable multi-agent orchestration with retries and no timeouts. Works with AI SDK and LangChain tools.
  </Card>
  <Card title="Box" icon="cube" href="/box/overall/quickstart">
    Sandboxed cloud containers with Claude Code, Codex, or OpenCode built in, for running agents at scale.
  </Card>
  <Card title="Search" icon="magnifying-glass" href="/search/overall/getstarted">
    Full-text and semantic search over your documents, ready to serve as a retrieval tool.
  </Card>
  <Card title="QStash for LLM calls" icon="clock" href="/qstash/integrations/llm">
    Call LLM APIs with retries, callbacks, batching, and a 2-hour timeout from serverless runtimes.
  </Card>
</CardGroup>