Preface

When integrating agents with Upstash’s Redis, vector database, message queues, or workflows in DSH, the common practice is to repeatedly paste documentation snippets or manually explain SDK usage in conversations. With documentation scattered across multiple product lines, agents often confuse APIs, miss initialization steps, or choose the wrong integration method in scenarios like QStash or Workflow.

The following introduces upstash/skills. It organizes various Upstash product lines into a set of loadable Agent Skills, accompanied by MCP server configurations, allowing coding agents to read relevant instructions based on tasks rather than piecing together documentation from scratch.

What This Is

upstash/skills is a skill collection maintained by upstash, designed for AI coding agents using Upstash SDKs. The repository describes itself as “packaged instructions and resources that extend agent capabilities.”

The same repository supports distribution via Agent Skills, Claude Code plugins, Cursor plugins, OpenAI Codex plugins, and DeepSeek Harness bundles. It is classified as “networked tools” in the SkillHub community directory, currently with 21 GitHub stars and 7 forks.

Included Skills

The repository README lists the following sub-skills, each corresponding to an Upstash product or toolchain:

Skill Description
upstash Comprehensive skill covering all Upstash SDKs
upstash-box-js Sandbox cloud container with AI agent, shell, filesystem, and git (JavaScript)
upstash-box-py Same as above, but for Python SDK
upstash-cli Access Upstash Developer API via upstash CLI
upstash-qstash-js Serverless messaging and scheduling based on HTTP endpoints
upstash-ratelimit-js Rate limiting with Redis Rate Limit TypeScript SDK
upstash-redis-js Serverless Redis: caching, sessions, leaderboards, full-text search, etc.
upstash-redis-start Quickly provision temporary, zero-configuration Redis for agents without registration
upstash-search-js Introduction to full-text search, core concepts, and TypeScript SDK
upstash-vector-js Vector database features, SDK usage, and framework integration
upstash-workflow-js Persistent workflows: defining, triggering, and managing multi-step processes

Agents load relevant sub-skills based on tasks, avoiding dumping all documentation at once.

Installation in DeepSeek Harness

DSH installation writes both the skill package and Upstash MCP server, requiring pnpm in the system PATH.

1、Add the plugin to the specified profile (web is the profile used when starting dsh web):

dsh plugin --profile web add github:upstash/skills

2、Start the Harness:

dsh web

3、Enter Upstash credentials in the session (optional, only needed when MCP connects to an account):

/upstash-login YOUR_EMAIL YOUR_API_KEY

Credentials are saved in ~/.dsh/.credentials.yaml. The README states that skills can be used without credentials; once email and API Key are provided, the MCP server automatically connects for account-level operations like creating databases and viewing QStash logs.

Typical Usage

Query by Product Line

After installing and starting DSH, directly describe your goal in the conversation, such as “use upstash-redis-js for session caching” or “use QStash to schedule webhook calls.” Agents should load SDK instructions and examples from the corresponding sub-skill rather than reading the entire Upstash documentation.

Temporary Redis

When you need to quickly verify Redis logic without going through the full registration process, point to the upstash-redis-start skill and follow its steps to provision a scratch database.

Account-Level Operations Require MCP

Operations that call the Upstash API, such as creating production databases, publishing messages, or querying vectors, are handled by the bundled @upstash/mcp-server after login. Even without executing /upstash-login, skills can still provide code and concept guidance but cannot operate on your account resources.

Other Clients (Reference)

The same repository also supports the generic Agent Skills CLI:

npx skills add upstash/skills

Clients like OpenCode and Zed can install via the --agent parameter; Claude Code and Codex have their own independent marketplace commands. If you’re not working in the DSH environment, refer to the GitHub README for relevant sections.

Applicable Scenarios and Notes

Who It’s For

  • Full-stack or backend developers who are already using or planning to use Upstash Redis, Vector, Search, QStash, Workflow, Box, and other services;
  • Teams hoping to reduce the “look up docs—copy—modify parameters” cycle in coding agents like DSH, Cursor, and Claude Code.

Please Confirm Before Use

  • The plugin runs with the current DSH process permissions; before installation, it’s recommended to read the github.com/upstash/skills source code and license to ensure it meets your security and compliance requirements.
  • The Cursor official marketplace hasn’t yet listed this plugin (README says it’s awaiting listing); DSH users should use the dsh plugin --profile web add path described above, rather than assuming it can be installed with one click in the Cursor Marketplace.
  • When involving real accounts and data, API Keys should only be written to local credential files and not committed to version control.

Conclusion

upstash/skills packages Upstash’s multi-product SDKs and operational instructions into skills that can be loaded on demand, and installs them alongside the MCP server in DSH. It’s suitable for scenarios where agents assist in writing and debugging Upstash integration code.