CockroachDB and AI

On this page Carat arrow pointing down

CockroachDB supports AI in two primary ways: as a platform for AI-assisted development and as a data store for AI applications.

Support for AI-assisted development: CockroachDB integrates with AI development tools, including Claude Code, Cursor, and GitHub Copilot in VS Code. MCP servers provide access to CockroachDB Cloud clusters and CockroachDB documentation. The Agent Skills repository encodes operational workflows in a machine-executable format. The ccloud command-line interface (CLI) provides programmatic access to CockroachDB Cloud. These tools enable management, deployment, and development of CockroachDB clusters and applications using natural language prompts and AI-assisted workflows.

For more information, refer to Support for AI-assisted development.

A data store for AI applications: CockroachDB can serve as the system of record and the retrieval engine for AI applications. It combines native support for vector data and similarity search with strongly consistent transactions, horizontal scalability, and multi-region deployments. These capabilities support storing vector embeddings, agent state, conversation histories, and other AI-related data alongside relational data.

For more information, refer to CockroachDB as a data store for AI applicatons.

Support for AI-assisted development

CockroachDB enables your AI development tools to work directly with CockroachDB and CockroachDB Cloud. It supports:

The following sections provide an overview of these tools and how they support AI-assisted deployment and maintenance of CockroachDB clusters, and AI-assisted development of applications that are built on CockroachDB.

CockroachDB Cloud MCP server

The CockroachDB Cloud MCP server is a managed endpoint in CockroachDB Cloud that exposes a set of tools for inspecting and querying your clusters from your AI tools. These tools let your AI assistants list databases and tables, describe schemas and indexes, inspect cluster health and running queries, and run read-only SQL and EXPLAIN statements. When explicitly enabled, they can also create databases, create tables, and insert rows.

This access allows you to manage and modify a CockroachDB Cloud cluster using natural language prompts. For example, you could use your AI tool to do the following:

List all of the tables in the movr database.
If there are no tables, create one called "inventory" with columns for id, product_name, quantity, and last_updated.

Agent Skills for CockroachDB

Agent Skills for CockroachDB are small, structured capabilities that encode CockroachDB operational expertise in a machine-executable format. They live in the public cockroachdb-skills repository and follow the Agent Skills Specification, with defined inputs, outputs, and safety guardrails.

Each skill focuses on a specific task, such as auditing user privileges, triaging live SQL activity, validating production readiness, or checking backup and disaster recovery posture. Skills link back to the relevant CockroachDB documentation rather than duplicating it. They are organized into domains like onboarding and migrations, application development, performance and scaling, operations and lifecycle, resilience and disaster recovery, observability and diagnostics, security and governance, integrations, and cost management.

Your AI agents and IDE integrations can consume these skills directly, so you can reuse the same operational workflows across different toolchains.

ccloud command-line interface

The ccloud CLI is the command-line interface for CockroachDB Cloud. With a single binary, you can install ccloud and then create clusters, manage networking (for example, IP allowlists), create SQL users, and retrieve connection information. Commands such as ccloud quickstart, ccloud cluster create, ccloud cluster sql, and ccloud cluster delete cover the core lifecycle of CockroachDB Cloud clusters.

Because ccloud is text-based and follows a stable command structure, it works well as a surface for AI tools and automations. An AI assistant can generate or run ccloud commands to set up clusters, rotate credentials, or retrieve connection URLs, while you keep access mediated through the CLI and existing Cloud authentication.

CockroachDB Docs MCP server

The CockroachDB Docs MCP server exposes the published CockroachDB documentation to your MCP-compatible tools over HTTP. After you add the server configuration to your client, your AI assistant can answer questions using the official docs without leaving your editor—for example, how to configure multi-region clusters, tune queries, or set up TLS and RBAC.

This integration only provides access to CockroachDB product documentation. It does not connect to your clusters or data. It is intended to give your AI tools reliable product information while you develop and operate CockroachDB and CockroachDB Cloud. For example, you could use your AI tool to ask the following:

How do table statistics get refreshed in CockroachDB?

CockroachDB as a data store for AI applications

CockroachDB provides the database features needed to store and query AI-related data, including vector embeddings and agent state, with the same transactional guarantees as your other workloads.

Vector search and RAG

AI applications often represent text, images, and other content as vector embeddings. These are numerical representations that capture semantic meaning. To find relevant information, AI applications need to search for vectors that are similar to a query vector, typically using distance metrics. This similarity search is fundamental to retrieval-augmented generation (RAG), semantic search, and recommendation systems.

CockroachDB has a VECTOR data type for storing fixed-length floating-point embeddings and supports similarity operators such as L2 distance (<->), inner product (<#>), and cosine distance (<=>). You can index vectors using CREATE VECTOR INDEX and combine them with other indexed columns.

You can store vector embeddings, relational data, and JSON metadata in the same table and query them together. For example:

SELECT id, name
FROM items
WHERE customer_id = 1
ORDER BY embedding <-> $query_vector
LIMIT 5;

Retrieval-augmented generation (RAG) systems depend on fast and accurate retrieval of contextual data. CockroachDB's vector support enables you to implement RAG and semantic search patterns so that you can use CockroachDB as the data store for AI applications. Vector indexes enable efficient similarity searches over large datasets. You can store document embeddings in a cluster alongside the source documents, metadata, and access control information associated with those documents. When a user query comes in, you can use vector similarity operators to retrieve semantically relevant documents, filter by permissions, and return both the relevant context and any associated metadata in a single transaction.

See the RAG tutorial for a complete implementation example.

AI agent state and workflow coordination

AI agents that perform autonomous operations require durable storage for execution state, workflow metadata, and operational history. These agents must track state transitions across multi-step processes, coordinate activities between concurrent executions, and ensure that operations can be safely retried or resumed after failures.

CockroachDB's transactional model provides a foundation for storing agent state, execution history, and coordination metadata. Serializable isolation ensures that state transitions occur correctly even when multiple agents or processes attempt concurrent updates. The database's high availability design allows agents to continue operating during node or region failures without requiring external coordination services.

Scale, consistency, and governance

AI applications typically generate high data volumes, serve globally distributed users, and require both transactional correctness and operational durability. Conversation histories, vector embeddings, feature tables, and agent state accumulate quickly and are accessed across regions. These characteristics make AI workloads well-suited to CockroachDB's core design:

  • CockroachDB scales horizontally by adding nodes to increase capacity.
  • Data is automatically replicated and rebalanced across the cluster and node failures do not require application-level failover.
  • Multi-region deployments place data closer to users and can enforce data locality or residency requirements while maintaining strong consistency.

CockroachDB provides serializable transactions by default. Vector indexes participate in the same transaction and index maintenance model as other secondary indexes, so similarity search results remain aligned with the underlying data. This applies to tables storing embeddings, relational records, and agent state.

See also

On this page

Product

Resources

Learn

Support Channels

Company

Get developer news

×
★ ★ ★ ★ ★