Agent Plugins Overview

CloudPeek publishes two official Agent Plugins: portable packages that connect an external AI client, such as Claude Code, Cursor, GitHub Copilot or Codex, to CloudPeek's hosted services. Installing one gives that client a set of CloudPeek tools plus the workflow knowledge to use them properly.

They are the route in for anyone working outside the CloudPeek web app: an engineer reviewing a change in their editor, an analyst asking questions from a chat client, or an agent running unattended in a pipeline.

Both plugins live in the public Cloud-Peek/agent-plugin repository.

Which one do you want?

They do different jobs, connect to different services and authenticate differently, so install whichever suits the client you are setting up. Installing both is perfectly normal: an engineer might use Vulnerability Intelligence to understand a CVE and CloudPeek to check whether it has actually fired anything in their estate.

Each page above starts with the installation commands for every supported client, so if you already know which plugin you want, go straight there.

What a plugin actually contains

Each plugin root is a portable Agent Plugins 1.0.0 package:

• plugin.json — the manifest: name, version, description, author, licence.
• mcp.json — the MCP server the plugin connects to, over Streamable HTTP. Exactly one server per plugin, defined in exactly one place.
• skills/ — Markdown workflow instructions the agent loads on demand.

There is no build step and nothing to compile. Your client reads mcp.json, connects to the hosted endpoint, and loads the skills when a request matches one.

Tools come from the server, skills come from the package

This split is worth understanding, because it explains what changes when.

The tools live on the hosted MCP endpoint. When CloudPeek adds or changes a tool, your client picks it up on its next connection: there is nothing to reinstall.

The skills ship inside the plugin package. They are the workflow knowledge, which order to call tools in, what counts as enough evidence, what a useful answer looks like. Updating those means refreshing the plugin from the marketplace.

Compatible clients

Every client on the compatible clients list can load either plugin as-is, because both are plain spec-compliant packages: GitHub Copilot, VS Code, Cursor, ChatGPT and Codex, Kiro, Grok Bot, OpenClaw, NanoClaw and Hermes Agent, alongside Claude Code.

Some clients also read a catalogue manifest for presentation, giving them a display name, an icon and starter prompts. Those are shipped alongside the portable package as additions rather than replacements, so a client that reads none of them still installs the plugin from the root manifest. Every one of them points at the single mcp.json.

Any other MCP client works too, as long as it speaks the Streamable HTTP transport. The per-plugin pages give the raw configuration for that case.

Authentication, in one paragraph

The two plugins sit at opposite ends of the access spectrum, deliberately. Vulnerability Intelligence is anonymous: no account, no sign-in, no token to manage, and it holds nothing about you. CloudPeek reaches your tenant's real incident data, so it requires a sign-in, and what the agent can see and do is scoped to your CloudPeek user and enforced server-side. Full detail is on each plugin's page.

Where to go next

• CloudPeek Plugin, for incident triage and investigation.
• Vulnerability Intelligence Plugin, for CVE and CWE lookup.
• API Overview, the underlying REST API the CloudPeek gateway exposes.