MCP
A FinOps MCP server that optimizes Google Cloud CUDs
There are a growing number of FinOps MCP servers. Almost all of them do the same thing: let an AI assistant read and summarize your cloud bill. Solvicus does something different — it solves the commitment-purchase decision itself.
What most FinOps MCP servers do
The Model Context Protocol (MCP) is an open standard that lets AI assistants call external tools — see the FAQ for a primer. The FinOps servers built on it are, as of July 2026, overwhelmingly read-only reporting tools. Open-source examples like CloudScope MCP and focus-mcp expose your billing data for natural-language questions: what did we spend by service, where are the anomalies, which resources look idle. Cloud providers ship similar servers for querying spend.
They're useful, and deliberately safe — read-only by design. But they answer “what did we spend?” They don't answer “which commitments should we buy, for how long, to cover the demand we know is coming?”That's an optimization problem, not a query.
What Solvicus adds
Solvicus is an MCP server whose tool is the optimizer. You describe the demand you already know — signed contracts, a planned migration, a steady baseline — plus the commitments you already own and your Google Cloud pricing. It models each resource as a mixed-integer program, solves for the lowest-cost mix of 1- and 3-year committed use discounts, and returns the plan with an interactive artifact: coverage, utilization and cost charts, the reasoning behind each commitment, and editable purchase rows you can adjust to stress-test the plan.
It stays read-only where it matters: Solvicus has no access to your cloud account — no billing permissions, no console access. It recommends and evaluates; you make the purchases. The data it works from is what you or your assistant hand it.
Using it from your assistant
Solvicus runs as a remote MCP server, so setup is a server URL and a key or OAuth client — no local install. It works with Claude Desktop, ChatGPT Desktop and Antigravity; the setup steps for each client are on the homepage. From there the loop is conversational: describe your demand, get a commitment plan, adjust it, and ask follow-up questions against the same model.
Why conversational delivery matters here
Commitment planning is unusually input-heavy: contracts with start and end months, per-resource demand, existing CUDs and their expiries, current pricing. Assembling that by hand is the reason the decision often gets deferred. An AI assistant is genuinely good at that gathering-and-shaping work — pulling numbers out of a spreadsheet or a contract summary and turning them into model inputs — while the solver does the part language models are bad at: exact, reproducible optimization. Neither half is sufficient alone.
Learn more
- Google Cloud CUD optimization guide — coverage, utilization, break-even, and how the plan is computed.
- How GCP CUD tools compare — input models, automation, and pricing across the market.
- The CUD optimization solver — required inputs, outputs, and an example prompt.