MCP server
Query your ZopNight workspace from Claude, Cursor, Codex, or any MCP client; 43 read-only tools, PAT-authenticated, write-blocked at the protocol.
The ZopNight MCP server lets an AI assistant read your workspace directly; ask Claude “what are my top idle EC2 instances this week?” and it queries live resources, costs, and recommendations instead of guessing. It speaks the Model Context Protocol, so any MCP-capable client works.
What you can connect
- Claude Desktop
- Cursor
- Codex
- Claude Code
- Any other MCP-compatible client
What it exposes
43 read-only tools spanning eleven workspace domains:
resources · schedules · costs · recommendations · teams · budgets · audit logs · cloud accounts · overrides · notifications · billing sync.
List tools forward the full filter and pagination set (provider, status, region, search, sort, …), so the assistant can ask precise questions (“stopped RDS instances in eu-west-1 for account X”) rather than pulling everything and filtering client-side.
Authentication
The MCP server authenticates with a personal access token (zn_pat_…). Mint one from Settings → Access tokens (see Authentication).
- The token is validated at the gateway, and RBAC is enforced there; the assistant sees exactly what your role and scopes allow, nothing more.
- MCP access is controlled by an org-level toggle; an admin enables it for the workspace.
- Every MCP call is written to the audit trail (see RBAC & audit), so you can see what an assistant queried and when.
Setup
Enable MCP for the org
An admin turns on the org-level MCP toggle in workspace settings.
Mint a read-only PAT
Settings → Access tokens → create a token. It’s shown once; store it in your secret manager.
Point your client at the server
Add the ZopNight MCP server to your client’s MCP configuration, using the PAT as the credential. Your client’s docs cover where its MCP config lives.
Ask a question
From the assistant, ask something like “show me open recommendations over $500/month on AWS”; it calls the recommendations tool with the right filters and reads back the result.
Performance
Expensive aggregation tools (cost rollups, summaries) are served from a short-TTL in-memory cache, so repeated questions in a session stay fast without re-querying the whole estate each time.