Introducing Chainguard MCP for secure agentic development

AI coding tools and agents write code faster than ever, and the bottleneck in the SDLC is shifting downstream. Review cycles are quickly becoming the new constraint. 

Agents are great at writing code, but they aren’t great at knowing what’s trustworthy. Pulling unverified dependencies into codebases expands the enterprise attack surface and creates major supply chain risk that’s difficult to review manually. Engineering leaders need a way to scale agentic velocity without compromising security.

Today, we’re announcing four Chainguard MCPs that help provide a secure foundation for teams to ship AI-generated code to production, at machine speed. 

How Chainguard MCP works

Chainguard MCP delivers four supported servers to secure every stage of dependency selection.

• The cg-oci server confirms an image tag or digest exists before it lands in a Dockerfile, preventing agents from hallucinating wrong tags.

• The cg-apk server looks up exact Chainguard package equivalents by name, eliminating guesswork around package naming.

• The cg-versions server checks whether a package version is current or end of life before an agent recommends it.

• The cg-api server verifies organization policy, entitlements, and attestation data programmatically.

Every artifact comes backed by cryptographic provenance and Software Bill of Materials metadata so agents verify security automatically.

Server-side controls enforce governance across any MCP client

Chainguard MCP fits cleanly into existing workflows. Instead of relying on prompt guardrails, governance is enforced on the server side. Response caps prevent unbounded queries, lower token consumption, and stop context window bloating during long coding sessions.

Chainguard MCP works across any client supporting the Model Context Protocol using standard OAuth login flows. Engineering teams can now scale agentic velocity with complete confidence in their software supply chain.

Chainguard MCP is now generally available for enterprise customers 

Enterprise customers can start using the Chainguard MCP today. Each of the four servers, cg-oci, cg-apk, cg-versions, and cg-api, is reachable by any MCP-compatible client (Claude Code, Cursor, and others) using a standard OAuth login.

To connect:

1. Add the server to your MCP client using its URL, the same way you'd add any other MCP server.

2. On first connection, your client will prompt a standard browser-based OAuth login. Approve it, and the server is ready to use.

3. Prefer the CLI, or working in a headless or remote environment? Log in with chainctl auth login --audience=<server-url>, once per server, and your client will authenticate automatically on future connections with no repeated browser prompts.

For full setup instructions, available tools, and configuration examples for each server, see the Chainguard MCP docs.