# Finout MCP Integration (Beta)
### Overview
With the Finout MCP ([Model Context Protocol](https://modelcontextprotocol.io/introduction)), you can ask questions about your cloud cost spend, anomalies, waste, budgets, tag coverage, unit economics, and more - directly inside the AI tools you already use. The Finout MCP acts as a bridge to your live Finout data, letting you query and visualize cloud spend through natural language without switching tools or exporting data.
The Finout MCP is hosted and managed by Finout. There is nothing to install and no credentials to manage locally - authentication is handled via a standard OAuth flow, the same experience you use to log in to Finout.
### MCP Clients
An MCP client is any AI assistant or agent that can communicate using the Model Context Protocol. The Finout MCP is compatible with any client that supports remote MCP with OAuth. Examples include:
* [Claude](https://claude.ai/) (web and desktop)
* [Cursor](https://www.cursor.com/) IDE
* Any other MCP-compatible AI tool
You can find a full list of clients in the[ MCP documentation](https://modelcontextprotocol.io/clients).
### What You Can Do
Once connected, ask questions like:
* "Show me the top 5 AWS services by cost last month"
* "Compare our cloud spend this quarter vs last quarter"
* "Which dimensions had the biggest cost increase this month?"
* "Are there any high-severity cost anomalies this week?"
* "What are our biggest waste and rightsizing opportunities?"
* "How are we tracking against our Q1 budget?"
* "What percentage of our spend is tagged with a team tag?"
### Getting Connected
#### Step 1 - Download and Install an MCP Client
* Claude - Download[ Claude for Desktop](https://claude.ai/download) or use[ Claude on the web](https://claude.ai/).
* Cursor - Download[ Cursor](https://www.cursor.com/) for your operating system.
Other tools - Any MCP-compatible AI tool that supports remote connections via OAuth will work. Point it to .
#### Step 2 - Configure the Finout MCP Server
Claude
1. Open Settings → Connectors
2. Click Add Custom Connector
3. Enter the following details and click Add:
| Field | Value |
| ----- | --------------------------- |
| Name | Finout |
| URL | |
Cursor
1. Press Cmd/Ctrl + Shift + P to open the command palette
2. Type MCP and select Open MCP Settings
3. Add the following to your mcp.json file:
{
"mcpServers": {
"finout": {
"url": ""
}
}
}
4. Save the file
#### Step 3 - Authorize the Finout MCP Server
After configuring the server, your browser will open an authorization screen.
1. Review the authorization details and click Allow Access
2. Log in to Finout with your existing credentials or via SSO
3. Once authorized, close the browser tab and return to your AI client
Tip: By default, most clients ask for approval before each tool call. For a smoother experience, set the permission to Always Allow in your connector settings.
#### Step 4 - Start Prompting
Open a new chat and start asking questions about your cloud costs. For example:
"Show me the top 5 AWS services by cost last month"
Your AI client will use the appropriate Finout MCP tool and return results from your live Finout data.
The Finout MCP exposes 25 tools across five categories.
#### Discovery & Context
| Tool | Description |
| ------------------------ | ---------------------------------------------------------------------------------------------------------------------------------- |
| get\_account\_context | Returns connected cost centers, available data ranges, and account configuration. Good first call when orienting to a new account. |
| search\_filters | Finds filter metadata by name - required before calling query\_costs or compare\_costs. |
| get\_filter\_values | Returns available values for a specific filter (e.g., "what services do we use?"). |
| list\_available\_filters | Lists all available filters by cost center. Use only when you need a full overview - prefer search\_filters for targeted lookups. |
| debug\_filters | Diagnostic tool for inspecting raw filter metadata. Use when a filter lookup returns unexpected results. |
| discover\_context | Finds dashboards, saved views, and data explorers by name. |
| list\_data\_explorers | Lists all saved data explorer configurations in the account. |
| list\_telemetry\_centers | Lists custom metric sources available for unit economics calculations. |
#### Cost Analysis
| Tool | Description |
| ----------------------- | ---------------------------------------------------------------------------------------------------------------------------------- |
| query\_costs | Core cost query tool. Returns totals, breakdowns, and trends for any combination of time period, filters, and grouping dimensions. |
| compare\_costs | Compares cloud costs between two time periods. |
| get\_top\_movers | Identifies dimensions with the biggest cost changes between two periods. |
| get\_unit\_economics | Computes cost-per-unit metrics (e.g., cost per active user, cost per API request). |
| get\_cost\_statistics | Returns daily cost stats: mean, median, peak, trough, and volatility. |
| get\_cost\_patterns | Analyzes temporal patterns - hourly peaks, weekday vs. weekend splits, recurring cycles. |
| get\_savings\_coverage | Analyzes savings plan and reservation coverage, including commitment gaps. |
| get\_usage\_unit\_types | Discovers available usage unit types for a cost center. Call before using usage\_configuration in query\_costs. |
#### Anomalies & Waste
| Tool | Description |
| --------------------------- | ------------------------------------------------------------------------------------------------------------ |
| get\_anomalies | Retrieves cost anomalies and spikes detected by Finout. Filter by severity: high, medium, low. |
| get\_waste\_recommendations | Returns CostGuard waste detection results - idle resources, over-provisioned instances, and commitment gaps. |
#### Governance & Allocation
| Tool | Description |
| ---------------------- | ------------------------------------------------------------------------------------------ |
| get\_tag\_coverage | Measures what percentage of spend is tagged by a given dimension. |
| get\_financial\_plans | Retrieves budgets and forecasts with actuals, run rate, and forecast per dimension. |
| analyze\_virtual\_tags | Deep-dives into virtual tag configuration, relationships, and allocation logic. |
| get\_object\_usages | Finds all places a named Finout object is referenced - useful before modifying anything. |
| check\_delete\_safety | Checks whether a Finout object is safe to delete without breaking downstream dependencies. |
#### Presentation
| Tool | Description |
| ------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------- |
| render\_chart | Renders a bar, column, line, or pie chart directly in the AI assistant UI. Use after query\_costs or compare\_costs to visualize results in-context. |
### Prompting Best Practices
Add context to your prompts. Mention the provider, time period, service, or tag dimension when relevant. For example: "Break down EC2 costs by environment tag for the last 30 days"
Be specific about the time period. The default is the last 30 days - saying "last quarter" or "March 2026" ensures you get exactly the window you intend.
Ask one thing at a time. Stacking unrelated questions can cause your AI client to conflate results. Ask them separately for cleaner answers.
Specify the output format. If you want a chart, table, or summary, say so. For example: "Show me monthly EC2 spend by region for Q1 as a bar chart"
#### Validating Responses
Because the MCP relies on an LLM, responses are not always deterministic. To validate outputs:
* Cross-reference results against your existing Finout dashboards or saved data explorers
* Expand the tool calls in your AI client to inspect the underlying filters and groupings applied
* If a response looks off, note what you asked, what you received, and what you expected - and share it with us via the submit\_feedback tool
### Feedback
The Finout MCP includes a built-in submit\_feedback tool. After any interaction, ask your AI client to submit feedback directly: "Please submit feedback: the get\_top\_movers tool returned data for the wrong time period when I asked about last quarter."
You can also reach your Finout contact directly or email . The more specific your feedback, the faster we can act on it.
Need help? Reach out to