Automation — the AI / Claude Code layer
Clustersco's tooling thinks and acts through a small, shared operating layer: skills we develop,
MCP servers we connect, and agents we define. It's distributed as Claude Code plugins
through the clustersco marketplace, so the whole team installs the same tooling with one command.
The marketplace
/plugin marketplace add https://bitbucket.org/clustersco/<org-repo>.git # once
| Plugin | What it adds |
|---|---|
microsoft-entra | Read-only Microsoft 365 / Entra directory access (MCP) |
clustersco-standards | Engineering-standards + scaffolding skills (architecture, git, governance, design, playbooks) |
Principles
- Portable — a skill or agent authored once drops into any repo unchanged, the same way a module drops into any app.
- No secrets in config —
.mcp.jsoncarries only IDs and OAuth config, never client secrets. - Read-first — MCP servers use read-only scopes unless a workflow demonstrably needs write.
- Runbooks, not tribal knowledge — every server and skill ships with how-to + verify docs.