Skip to content
Cloud Sleuth
Get early access

Capability 07 · AI agents

MCP Server for AI Agents

Give Claude, Cursor and your own AI agents read-only answers about your AWS and Azure estate: who owns a resource, what depends on it and what's unowned, before they act.

? Answers: Can my agent act on it safely?

Get early access CLI vs. Cloud pricing
Cloud Sleuth CLI
A local MCP server that runs with the CLI, so AI tools on your machine can query your own inventory and owners.
Cloud Sleuth Cloud
A hosted MCP endpoint across every account, with per-agent permissions and an audit log of every question asked.
Roadmap
Phase 1: local MCP server · Phase 2: hosted MCP endpoint with per-agent permissions

More and more teams are connecting AI agents to their cloud accounts through the Model Context Protocol (MCP). Agents can already list resources and call cloud APIs. What they can’t see is the context a careful engineer checks first: who owns this resource, what depends on it, and whether anyone would notice if it disappeared. An agent asked to “clean up staging” has no way to know that the quiet VPC it found carries egress for three teams.

The Cloud Sleuth MCP server gives agents that context, read-only, from the same inventory and ownership model as the rest of Cloud Sleuth.

How it works

  • Read-only tools. The server exposes tools such as who_owns, blast_radius, find_unowned and query_inventory. None of them can change your infrastructure.
  • Answers with evidence. Each answer comes with the reasons behind it, like the CloudTrail creator, the IaC repository or the resources that depend on it, so the agent and the person reviewing its work can check it.
  • Works with the tools you use. Any MCP client can connect, including Claude, Cursor and in-house agents.
  • Local first. With the CLI, the server runs on your machine against your own inventory, and nothing leaves your environment.
  • Permissions and audit. In Cloud Sleuth Cloud, a hosted endpoint covers every account, each agent gets its own scoped permissions, and every question is logged.
  • A check before acting. Point your agent’s instructions at blast_radius before any destructive change, so it asks before it deletes.

Illustrative example - tool names and output not final:

tool   blast_radius("vpc shared-egress")
owner  platform (tag)
used   payments, data, ci · 2 NAT gateways · 14 ENIs
iac    terraform, repo infra-core
safe   no: shared resource with 3 dependent teams

Why it matters

AI agents make cloud changes faster, including the wrong ones. Ownership and dependency context is what separates a confident change from a guess, for people and agents alike. Giving agents the same answers your engineers get, read-only and with the evidence, makes automation safer without slowing it down.

Cloud Sleuth CLI vs. Cloud Sleuth Cloud

The Cloud Sleuth CLI includes a local MCP server, so AI tools on your machine can query your own inventory and owners. Cloud Sleuth Cloud is planned to add a hosted MCP endpoint across every account, with per-agent permissions and an audit log.

Both are pre-launch. Get early access to hear first when they are ready.