Today we are launching RisingWave Cloud V2, a ground-up rewrite of the RisingWave Cloud control plane built for how developers actually work in 2026. The streaming engine you already know is unchanged. What is new is everything around it: a redesigned console, a first-class command-line interface called rwc, and native support for AI coding agents through MCP and Agent Skills. If your database is going to be operated by software as much as by humans, it had better speak the language of both.
This post explains what we rebuilt, why we rebuilt it, and what you can do with it on day one.
Why now
A year ago, "AI in the developer workflow" mostly meant autocomplete. Today it means agents. Engineers we talk to are writing pipelines with Claude Code, reviewing PRs with Copilot, debugging incidents with Cursor, and shipping production SQL written half by them and half by a model. The agent is the new IDE.
That shift exposes a gap. Most managed databases were designed assuming a human in a browser tab. The console is the source of truth, the docs are a separate site, and the CLI, if one exists, is an afterthought. An agent landing in that environment is blind. It cannot see your clusters, it cannot run a migration, it cannot tell whether the materialized view it just suggested already exists, and it has no idea which streaming SQL patterns are correct for your database versus the one it saw last in its training data.
Cloud V2 is our answer. We rebuilt the control plane to be navigable, scriptable, and inspectable by software, not just by people. Humans get a faster console. Agents get an API, a CLI, an MCP server, and a curated knowledge pack. Everyone wins.
What we rebuilt
V1 of RisingWave Cloud served us well for several years and several thousand clusters, but it had accumulated the usual debts of any first-generation control plane. Page loads were slow. Multi-cluster workflows required a lot of clicking. Observability lived in a different mental model from cluster management. Most importantly, the platform was hard to drive from anywhere other than the console.
V2 is a clean break:
- A redesigned console. Faster navigation, clearer cluster-level views, a unified observability surface, and a sane information hierarchy for teams running more than one project.
- A redesigned control-plane API. Cleaner endpoints, consistent semantics, and a contract that the CLI and our internal tooling now share. The console is just one client of this API. The CLI is another. So is your CI pipeline.
- Better multi-cluster and multi-project management. Switching contexts, comparing clusters, and managing access across projects is no longer a multi-page journey.
- BYOC parity. Bring Your Own Cloud deployments are managed through the same surfaces as our hosted clusters. No second-class workflow.
We took the opportunity to question every assumption we had baked into V1. The result is a platform that is faster to use, easier to automate, and ready for what is coming next.
Meet rwc
The new RisingWave Cloud CLI is called rwc. It is the tool we wished we had been shipping all along.
rwc manages clusters, authentication, projects, and BYOC environments from your terminal. It is the same surface our own engineers use internally, which means it gets the same care and the same bug fixes. You can drive it from your laptop, from a CI job, from a Makefile, or from a shell script your coworker wrote at 2 a.m. and forgot about.
A few things rwc does today:
- Create, list, scale, and delete clusters.
- Manage authentication and projects.
- Configure and apply BYOC deployments on AWS, GCP, and Azure.
- Take and restore snapshots.
- Install agent skills into your local AI coding tools with a single command.
That last bullet is not a typo. The CLI is also our distribution channel for the AI-native pieces of Cloud V2. Which brings us to the interesting part.
Agent-native by default
Cloud V2 ships with three open-source projects that, together, make RisingWave the first managed streaming database designed to be operated by AI coding agents as a first-class user.
1. The MCP Server. risingwave-mcp is a Model Context Protocol server that connects your AI assistant directly to a running RisingWave instance. It exposes more than 100 tools covering query execution, schema inspection, DDL, source and sink configuration, streaming job monitoring (fragments, actors, backfill progress), storage analysis, and query plan inspection via EXPLAIN. It works with Claude Desktop, VS Code Copilot, and any other MCP-compatible client, over STDIO or HTTP.
In practice, this means an agent can answer questions like "which materialized views in this database depend on the orders source?" or "show me the backfill progress for this MV" without you copy-pasting SQL into a chat window. The agent sees the live database, not a stale screenshot of it.
2. Agent Skills. risingwave-agent-skills is a portable knowledge pack that teaches coding agents how to write correct RisingWave SQL. It follows the open Agent Skills specification, which means a single set of SKILL.md files works across Claude Code, GitHub Copilot, Cursor, Windsurf, Gemini CLI, Cline, and 18+ other agents we have tested.
The current skills cover two areas: a core risingwave skill (connections, MCP setup, sources, sinks, materialized views, time-windowed aggregations, watermarks, system catalogs) and a risingwave-best-practices skill with 14 rules covering schema design, materialized view patterns, CDC setup, sink configuration, and performance. These are the rules our own field engineers reach for. Now your agent reaches for them too.
3. The CLI as a delivery vehicle. rwc skill install --target claude-code installs the skills into your local Claude Code setup. The same command works for Cursor, Copilot, and the other supported agents. No repo cloning, no manual editing of configuration files. The CLI is also the only one of the three tools that requires a Cloud account. The MCP server and the Agent Skills are fully open source and work against any RisingWave deployment, hosted or self-managed.
What does this look like end to end? Imagine you ask Claude Code to build a real-time fraud detection pipeline on top of a Kafka topic. With the skills installed, the agent knows to use CREATE SOURCE with the right connector options, not a generic Postgres pattern. It knows when to use EMIT ON WINDOW CLOSE. It knows that materialized views in RisingWave are incrementally maintained, so the right answer is often an MV rather than a query. And with the MCP server connected, it can verify its assumptions against your live cluster before it writes a line of SQL. The difference between "agent that has heard of RisingWave" and "agent that actually knows RisingWave" is exactly the difference between a frustrating afternoon and a working pipeline by lunch.
What is staying the same
We are deliberate about what V2 does not change. The streaming engine, the SQL surface, the connectors, the Postgres wire compatibility, the disaggregated storage architecture on S3, the open-source license: none of that is moving. RisingWave is still a PostgreSQL-compatible streaming database written in Rust, still Apache 2.0, still the same engine that runs Nexmark faster than Flink on the majority of queries.
V2 is a control plane story. It is about how you reach the engine, not about the engine itself. If you have an existing cluster on Cloud V1, your data and your pipelines are safe, and migration is a managed process. We will reach out with details.
Getting started
Three commands get you most of the way:
# Install the CLI
brew install risingwavelabs/tap/rwc
# Log in and create a cluster
rwc auth login
rwc cluster create my-first-cluster
# Install agent skills into Claude Code (or cursor, copilot, etc.)
rwc skill install --target claude-code
From there, head to the Cloud V2 console for the full UI, the docs for deeper guides, and the GitHub repos for risingwave-mcp and agent-skills if you want to read the source or contribute.
What is next
This is a Public Preview. There are rough edges, and we will be moving fast. A few things we are actively working on:
- More skills, especially around CDC patterns, sink tuning, and incident response.
- Deeper MCP integration with the control plane itself, so agents can manage clusters and not just query them.
- Per-environment policies for what agents are allowed to do in production. The flip side of giving agents power is giving operators control.
- More CLI surface area, including streaming job introspection and lineage.
We built Cloud V2 because we believe the next decade of database operations will be co-piloted by software. The teams that win will be the ones whose infrastructure is legible to both humans and agents, with the same fidelity, through the same surfaces. That is the bet behind every line of code we shipped today.
Try it, break it, tell us what is missing. We are listening on GitHub, in our Slack community, and through every issue you open against the new repos. Thank you for building with us.
FAQ
Is RisingWave Cloud V1 going away? Not immediately. Existing V1 clusters continue to run. We will contact V1 customers directly with a migration plan. V2 is the recommended path for all new clusters.
Do I need a Cloud account to use the MCP server or agent skills?
No. Both risingwave-mcp and agent-skills are open source and work against any RisingWave instance, including self-hosted ones. Only the rwc CLI requires a Cloud account, and only for Cloud-specific commands.
Which AI assistants are supported? Agent Skills work across Claude Code, GitHub Copilot, Cursor, Windsurf, Gemini CLI, Cline, and more than a dozen others. The MCP server is tested with Claude Desktop and VS Code Copilot, and works with any MCP-compatible client over STDIO or HTTP.
Did the streaming engine change? No. V2 is a control plane release. The engine, SQL surface, connectors, and storage layer are unchanged.
Is the new console available to everyone today? Yes. Cloud V2 is in Public Preview and available to all new sign-ups. Existing customers will see the new console roll out over the coming weeks.
