Today Cockroach Labs is announcing new capabilities that make CockroachDB agent-ready, giving AI agents a secure, structured way to work with your database. These include a fully managed MCP Server, a redesigned ccloud CLI, and a public library of CockroachDB Agent Skills.
For most of the history of software, there has been one type of end user worth designing for: a human. A developer writing a query, a DBA tuning an index, an analyst building a report. However, that assumption is changing fast.
AI agents are increasingly as important as human users for many software products, and databases are no exception. These agents are already reading docs, writing code, provisioning clusters, triaging incidents, and generating more traffic than many human teams combined. The person in the loop is increasingly a supervisor of agents. We have been thinking hard about what that shift means for CockroachDB. We asked ourselves: “What does it take to support AI agents as first-class users?” If agents are reading schemas, modifying data, and operating clusters, they need more than access. Their requirements extend to structured interfaces, scoped permissions, and full auditability. This shaped three powerful features we are releasing today:
Together, they make CockroachDB agent-ready across the full database lifecycle.
How AI agents use databases: Application vs. operations
Agents interact with CockroachDB in two major ways.
Application backend – Agentic systems scale fast, their users expect always-on availability, and they operate across regions with tight latency requirements. CockroachDB handles this well by design: distributed data, horizontal scale without manual partitioning, and multi-region deployments that keep globally distributed workloads consistent. If you are building an agent-driven application and need a database that grows with your agent fleet without re-architecting, CockroachDB was already a strong answer.
Interaction surface – Agents are increasingly helping teams with the full database lifecycle: spinning up clusters, reviewing schemas, diagnosing slow queries, triaging alerts, coordinating upgrades. While there are humans in the loop in these workflows, the primary interaction is agent-driven.
Most of the industry conversation about databases and AI is focused on the first pattern. What we are announcing today specifically addresses the second, providing essential agent interaction capabilities.
What does “agent-ready” mean for databases?
An agent-ready database is a database designed for AI agents to safely and reliably interact with production systems – not just read data, but operate infrastructure. This calls for interfaces and controls that agents can reason over programmatically, while ensuring humans retain visibility and oversight.
In practice, that includes:
Structured interfaces that AI agents can reliably interpret and execute
Scoped permissions that govern exactly what actions an agent can take
Auditability so every action is logged and traceable
Interfaces that are model-agnostic, so it works across frameworks as your stack evolves
We built the MCP Server, ccloud CLI, and Agent Skills to provide these capabilities across different agent workflows.
Managed MCP Server: built for enterprise agents
The Managed MCP Server provides a secure, centralized interface for AI agents to interact with CockroachDB in enterprise environments. It’s designed for shared agent systems that operate across teams, services, and production workflows where access control, auditability, and governance are critical.
While a developer using Claude Code on their laptop is a single-player workflow, enterprise systems need stronger guarantees. Internal platforms such as observability tools, deployment systems, or data services often rely on agents that act on behalf of many users. In these environments, every action must be authenticated, authorized, and traceable.
CockroachDB Cloud’s MCP Server delivers these critical capabilities:
Authentication that integrates with your identity provider
RBAC that controls what each agent can do
Read/write consent enforced at the platform level
Audit logging that flows into your existing observability stack
Our fully managed MCP server is hosted, maintained, and secured by Cockroach Labs, supporting OAuth 2.0 for interactive workflows and service account API keys for autonomous agents in pipelines. Agents operate in read-only mode by default to explore schemas, inspect query plans, and run analytical queries. Any write operation requires explicit consent, and system tables are deny-listed. Every interaction is logged with tool name, and cluster context enabling production level usage for enterprises.
Because it is managed, there is no infrastructure to deploy or manage. A single configuration snippet from the Cloud Console connects Claude Code, Cursor, or any MCP-compatible client to a CockroachDB cluster.
For the architecture details and a closer look at how the authentication and consent model works, read our MCP Server post.
ccloud CLI: built for developer agents
We redesigned the ccloud CLI with agents as a first-class user. Developer agents like Claude Code, Codex, and Gemini CLI operate through shell commands, making the CLI the most natural interface for agents running on a laptop or inside CI/CD pipelines. CockroachDB’s ccloud CLI:
Requires no infrastructure to deploy
Works with any agent framework
Integrates cleanly with existing developer workflows
The interface is designed to be reliable for machines: commands follow a consistent structure, outputs are available as structured JSON, and errors are deterministic so agents can react programmatically. The CLI exposes the full CockroachDB Cloud API surface: from provisioning clusters to managing backups, networking, and replication.
Security is built into the model. Each agent operates under its own scoped service account, so access is bounded by permissions rather than prompts. When agents are operating in production, the permission system is the guardrail.
To learn more about our ccloud CLI, including how ccloud fits into CI/CD workflows and why CLI and MCP serve markedly different purposes, read the ccloud CLI post.
Agent skills: Built-in CockroachDB expertise for AI agents
The CockroachDB Agent Skills are a layer of domain knowledge that makes any general purpose LLM an expert . The CLI and MCP provide access, skills provide the context and judgment needed to use that access effectively.
Without Skills, an agent querying background system tables returns raw data. With Skills, it can diagnose issues, prioritize the right signals, and recommend next steps. For example, during a CPU spike, an agent can identify relevant background jobs, recognize what backups overlap with peak write traffic, and determine which schema changes to investigate.
Agent Skills are structured, machine-executable instructions that encode CockroachDB best practices across the full product lifecycle, including:
Onboarding and migrating data to CockroachDB
Query and schema design
Operations and management of the database
Performance and scaling
Security
Observability
And more
They follow open standards and work with any agent framework or LLM, whether the agent interacts with CockroachDB through the CLI or MCP Server.
We are publishing the CockroachDB Agent Skills repo openly so teams can start using them today. We’ll continue to update these Skills with every new version of CockroachDB.
For a step-by-step walkthrough of how skills work in a real incident, from a CPU spike alert all the way through diagnosis and recommended fix, read our Agent Skills post.
How MCP, CLI, and Agent Skills work together
There is a lot of ongoing debate about MCP versus CLI for agent access. That is the wrong question. MCP and CLI serve different scenarios, and the answer depends on who is using the agent and how. For example, a developer agent running locally needs the CLI, whereas a centralized enterprise system serving many teams needs MCP. Skills sit on top of both the CLI and MCP, bringing the same CockroachDB expertise regardless of which interface the agent is using.
We shipped all three because your database should not have an opinion about how the agent asks. Your agent-ready data architecture needs a secure enterprise-grade MCP for context and governance, CLI for control and speed, and Skills for better reasoning and workflows. Together they allow AI agents to become first-class users of the database.
Why databases must be built for AI agents today
The shift to agent-assisted development and operations is already happening in production and infrastructure expectations are changing with it. Databases that work well with agents expose their internal state clearly, while supporting structured and scoped access. They also give teams the confidence that what an agent does can be tracked and, if necessary, reversed.
CockroachDB was built with those properties already in place. The MCP Server, CLI redesign, and Agent Skills make them accessible to agents in a consistent, safe, and auditable manner.
Cockroach Labs is early in empowering enterprises with AI agent readiness: The agent ecosystem is moving fast, and so are we. If you are already working with AI agents and CockroachDB, we would love to hear how you are using them.
Get started with the Managed MCP Server | Explore Agent Skills | Try ccloud CLI
Try CockroachDB Today
Spin up your first CockroachDB Cloud cluster in minutes. Start with $400 in free credits. Or get a free 30-day trial of CockroachDB Enterprise on self-hosted environments.
Lakshmi Kannan leads our Product Management and Design function at Cockroach Labs.
