An MCP server for PruneOps: FinOps data for any agent

PruneOps now offers a way to access its data without opening a browser: an MCP (Model Context Protocol) server that any standard-compatible agent can connect to directly.

What MCP is, and why it's there

MCP is the open standard that lets an AI agent discover and call tools exposed by an external service in a structured way. Instead of scraping a UI or maintaining a custom integration for every tool, the agent connects to an endpoint, lists the tools available, and calls them with whatever parameters they need.

With PruneOps, that means any MCP client — Claude Desktop, Claude.ai, or your own client — can ask things like "how much is being paid on AWS this month?" or "what savings recommendations are active?", and the agent queries the organization's real data in PruneOps to answer.

How authorization works

There's no shared token handed out manually. The server implements OAuth 2.1 with PKCE and Dynamic Client Registration (RFC 7591): the MCP client registers itself automatically the first time, then opens a standard authorization flow where you sign in and choose which organization the connection is allowed to query.

Once authorized, the client can call tools/list to discover what's available and tools/call to request data. If you manage more than one organization, there's a switch_organization tool to change context without re-authorizing.

What you can query

There are currently seven tools, all read-only:

• list_organizations — the organizations you have access to.
• list_accounts — connected cloud accounts.
• list_k8s_clusters — connected Kubernetes clusters.
• list_resources — resource inventory, filterable by type and account.
• get_costs — costs, per account or aggregated.
• get_recommendations — active savings recommendations.
• get_savings — realized savings.

None of these tools can change anything — it's read-only access, meant for querying, not for taking action.

Why this matters

More and more tools and workflows go through an MCP client as their entry point. Having this server query infrastructure cost data directly, with the same rigor (real metrics, not estimates) PruneOps already applies in the web UI, means FinOps information stops living only inside a dashboard and becomes available from any standard-compatible client.

Full documentation at www.pruneops.run/docs/integrations/mcp.