Skip to content
CloudOptifyDocs

MCP / AI assistants

CloudOptify speaks the Model Context Protocol (MCP) — the open standard that lets AI assistants use external tools. Connect a compatible assistant and it can answer cost questions straight from your live data.

What you can ask

Section titled “What you can ask”

Capability-level examples:

  • “What did the Machine Learning & AI team cost this month?”
  • “Where are our biggest savings opportunities right now?”
  • “Is any budget projected to be exceeded?”
  • “Compare this month’s spend to last month and explain the difference.”

The assistant reads the same data the portal shows — spend, groups, findings, recommendations, forecasts, alerts — and can run the same kind of read-only queries.

Acting on commands

Section titled “Acting on commands”

With elevated scopes on its token, the assistant can also manage your FinOps workflow, not just report on it:

  • write — create and update budget alerts, scheduled reports, and cloud groups, or acknowledge findings: “Create a budget alert for Platform Engineering — warn at 80% of €5,000 per month.”
  • run — trigger scans: “Run a fresh scan and summarize anything new.”

Grant scopes deliberately: a read-only token is enough for questions, and an assistant can never exceed the scopes on its token (or the role of the member who created it).

Connecting an assistant

Section titled “Connecting an assistant”
  1. In the portal, open API Access and mint an API token (a read token is enough for Q&A).
  2. In your MCP-compatible assistant, add CloudOptify as an MCP server using the endpoint shown on the API Access page and the token as the credential.
  3. Ask a question — the assistant discovers the available capabilities automatically.

Delegated setups (the assistant acts as the signed-in user) are supported through the standard OAuth authorization-code flow with dynamic client registration.

Governance

Section titled “Governance”
  • Read-only by default — tokens scoped to read can query but never change anything.
  • Same rules as the portal — the assistant sees only your organization’s data, governed by the same roles, scopes, and rate limits as the REST API.
  • Revocable — revoke the token and the assistant’s access ends immediately.

Rate limits & quota

Section titled “Rate limits & quota”

MCP (and the REST API) are governed by two plan-based limits:

  • Per-minute rate limit — a burst guardrail applied per caller. Exceeding it returns 429 with a Retry-After header and error: "rate_limit_exceeded".
  • Monthly quota — total requests per organization per calendar month; it resets on the 1st. Exceeding it returns 429 with error: "quota_exceeded" and the reset date.

Both limits depend on your plan; higher tiers get higher limits, and Enterprise is effectively unlimited (fair use). Every response includes the current state in headers so your client can self-pace:

X-RateLimit-Limit: 300
X-RateLimit-Remaining: 274
X-Quota-Limit: 1000000
X-Quota-Remaining: 838201