Equinix Agent Kit
Agent-native Fabric operations, packaged like developer infrastructure.
This local kit turns the demo into a believable product surface: MCP setup, account preflight, quote-aware pricing, Terraform review, and observability hooks in one installable agent workflow. It is meant to be reviewed in a single sitting — read this overview first, then jump to Quickstart when you are ready to wire it to a client.
Configure Claude Desktop, Cursor, or VS Code against the Fabric MCP endpoint.
Use governed componentsCompose pricing.md, preflight, tool trace, and Terraform PR components.
Run the workbenchGenerate a plan-only private AI inference topology from intent.
Design principles
- Every mutating MCP tool has a visible approval gate.
- Commercial estimates come from an LLM-readable pricing contract, not inferred prose.
- Terraform is a review artifact until account assets, permissions, and quote validation pass.
- Agent actions are attributable to a dedicated MCP user with scoped permissions.
- Trace data is exportable and survives long enough to support a security review.
Reference architecture
The agent flows from intent through a chain of read-only discovery, account preflight, and a Terraform review artifact. Apply only happens after a human approves the PR and the preflight passes. This is the loop the kit is shaped to run safely.
intent -> pricing.md contract -> Fabric MCP read tools -> account preflight -> Terraform PR -> human approval -> MCP create/update tools -> observability stream
What this product would add
Equinix already has the raw material: Fabric APIs, Terraform modules, MCP servers, portal context, and new Fabric Intelligence positioning. The Agent Kit packages those surfaces into a developer-grade operating model that feels closer to Stripe or Cloudflare: composable docs, safe defaults, copyable configuration, and account-aware actions.
The kit is not a replacement for the Equinix portal. It is a parallel surface for the platform engineers who would rather review a PR than click through a wizard, and for the agents those engineers want to delegate to.