Integrations

Connects to the tools your changes already flow through.

ChangeGuard AI reads change and environment state from where it already lives, and writes only with a Permit, where you asked it to. What each integration reads, writes and needs — and its claim class.

Fleetcentralized evidence plane — reads, never writes
Edgeoptional local evidence — never the execution path
Permitbounded authorization for one action
Executorgoverned action plane — Cloud where reachable, Local where required
IntegrationReadsWritesPermissionsEffortStatus
AWS / EKS — AWS Account Center & FleetEKS cluster discovery; workloads, events, quotas, posture — read-only, collected centrallyNothingOne discovery role (eks:ListClusters, eks:DescribeCluster) trusting ChangeGuard with your external ID; a per-cluster get/list grant you applyConnect AWS once; authorize clusters you selectLive
Cloud ExecutorNothing on its ownExactly one permitted patch per Permit, to workloads in your chosen namespacesOne execution role per environment with no AWS service permissions — it exists only to authenticate to EKS; Kubernetes authority is scoped separately through an EKS access entry and a namespaced get/list/patch RBAC grantGenerated per environment from its Execution section; nothing installedLive-proven
Local ExecutorNothing on its ownThe same single permitted patch, from inside your clusterIts own service account with a namespaced get/list/patch grant — never the Edge collector’s identityOne generated manifest, applied by youSupported
GitHub pull requests (OIDC)The changed manifests the workflow sendsA verdict on the pull request; the outcome after deploy, joined back to itThe short-lived identity token GitHub issues to the workflow — no API key, nothing to pasteConnect the repository; add the workflowLive
Any CI/CD or agent (GitHub Actions, GitLab CI, Jenkins, shell, coding agents)The change or manifest you submit for judgmentNothing in your environment; returns SHIP / HOLD / BLOCKA cicd-scoped API keyAdd a step or templateLive
Amazon EKS MCP (Kiro, Claude Code, Cursor and any MCP client)The exact manifest or resource body the agent is about to applyNothing in your environment; returns SHIP / HOLD / BLOCK, and a pre-tool hook admits the agent’s own write only on a fresh SHIP for that exact contentThe same cicd-scoped API key, through the CLI; ChangeGuard AI gains no EKS MCP permissionsOne installer, one hook (guide)Live
ChangeGuard Edge (optional collector)Node-local evidence (runtime signals, benchmarks) plus the same cluster stateNothing — Edge supplies evidence only and is never the execution pathRead-only ClusterRole. It lists Secret and RBAC objects to discover Helm and GitOps state; secret values are never sentOne generated installSupported
ArgoCD / Flux CDApplication and reconciliation state for GitOps change visibilityNothingRead access to Argo / Flux resources in-clusterAuto-discovered with EdgeSupported
GitHub App (fix pull requests)Repository metadata and contentsOpens pull requests with proposed fixes for you to reviewcontents:write, pull_requests:write, metadata:read on selected repositoriesInstall the appSupported
Slack / Microsoft Teams / webhookNothingOutbound notifications you configureAn incoming webhook URLPaste a webhookLive
Amazon Bedrock (analysis)——Runs in the ChangeGuard control plane, never in your environmentNoneLive
Self-hosted inference (NVIDIA NIM)——In-VPC inference for judgment and reasoning, for teams that require it—Roadmap

Claim classes: Live — running in production · Live-proven — exercised end to end in production acceptance and continuous security testing · Supported — available, used where the topology calls for it · Roadmap — planned, not shipping.

Judgment in CI, enforcement your call

Add a step and get SHIP / HOLD / BLOCK on the change. The shipped GitHub Action fails the job on BLOCK by default and never on HOLD; you can run it purely advisory. Whether a failed check stops a merge is your branch rule.

Execution only with a Permit

Reading your environment never grants writing — not through Fleet, not through Edge. The Executor acts only on a signed, single-use Permit bound to one action in one environment, whether it runs in ChangeGuard’s estate (Cloud) or in your cluster (Local).

Agents welcome

A coding agent that can change infrastructure gets the same judgment, the same Permit requirement and the same verified outcome as a person or a pipeline. One call to ask; nothing executes on its say-so. With the Amazon EKS MCP companion, an agent judges the exact change before it calls a privileged EKS MCP tool: AWS gives the agent the tool; ChangeGuard AI decides whether the exact change should happen.

Amazon EKS MCP

AI can make the change. ChangeGuard AI decides whether it should.

The EKS MCP companion puts the judgment immediately before an agent’s privileged write: a change judged SHIP is admitted, a scale-up judged BLOCK is denied, an environment never observed is HOLD, and a write with no judgment at all is denied. Setup takes one installer, one hook and one environment mapping; the steps are in the guide.

Watch: the Amazon EKS MCP companion on a live environment (2:53)
Terminal output is from a real companion run against Northwind, our live test environment; every verdict came from the live ChangeGuard AI backend. Works with Kiro, Claude Code, Cursor and any MCP client with a pre-tool hook.

Bring your stack

A first verdict on a pull request takes minutes to set up. Setup steps for every integration are in the docs.