/

FinOps MCP

FinOps MCP

FinOps MCP

An AI agent that runs your

FinOps

Costory understands your cost and its tech and business context, monitors and optimizes spend across hyperscalers, LLM providers and data clouds. Costory suggests, you decide in the conversation.

Works with Claude, Claude Code, Cursor, Grok Bot, GitHub Copilot, Dust and any MCP agent.

First optimization recommendation

The next day

ROI

Pays for itself with the first recommendation you act on

Setup

About 30 minutes

Data residency

EU

Access

Read-only on your cloud

Setup

Set up Costory

with one prompt.

Paste one prompt in your assistant. It signs you up and adds the Costory MCP, then shows you Costory on the demo dataset: a Team virtual dimension across AWS, GCP and Datadog, and a unit economics formula. Then it prepares the billing connection for you to apply. Under 30 minutes to connect, and your first recommendation the next day.

The Costory MCP is hosted and requires a Costory account. Signing in creates one with a demo dataset, so you can try team allocation and unit economics before connecting your billing.

Read-only access to billing exports and the metrics you connect. EU data residency.

Costory suggests, you decide in the conversation.

Try this prompt in Claude
Set up Costory for my company. If the Costory MCP isn't connected yet, tell me how to add it (https://app-api.costory.io/mcp). Signing in creates my Costory account with a demo dataset. If my workspace only has the demo dataset, start there and walk me through two use cases before we touch my billing. 1) Allocation by team: draft a Team virtual dimension that maps AWS, GCP and Datadog costs to teams across all three providers, preview the cost per team and the unallocated share, and explain which rule caught what. No need to publish. 2) Unit economics: build a formula like cost per daily active user from a usage metric in the demo, show the trend and split it by team. If the demo has no usage metric, tell me which one I would connect. Then suggest one more demo use case worth trying. Once I've seen the results, help me connect my real cloud or LLM billing with Terraform so the same team split and formula run on my data.
What Claude does next
01
Signs you in and adds the Costory MCP.
Creates your Costory organization with a demo dataset and connects your assistant to it.
02
Explores the demo with you.
Drafts a Team virtual dimension that splits AWS, GCP and Datadog costs by team, then builds a unit economics formula like cost per active user, split by team.
03
Connects your billing data.
Prepares the Terraform for your cloud or LLM billing export, so the same team split and formula run on your data. You review and apply it, nothing is installed in your infrastructure.
You approve what goes live. Costory never changes your cloud.
Open the Claude guide

Why it matters

From answering questions to running the loop

Raw billing data lets an assistant answer “how much”. Allocated cost plus tech context lets Costory’s agent work on its own: watch spend, explain changes, recommend what to optimize and follow up with the owner.

Optimize

What your assistant gets from Costory

Allocated cost by team, product, environment and feature

Deploys, releases and provider events on the same timeline

Service owners from your Virtual Dimensions

Usage metrics: CloudWatch, Cloud Monitoring, Datadog, your own metrics

Budgets, forecasts and alert history

With that, Claude Code proposes rightsizing, commitments, idle-resource cleanup, storage tiering and model routing changes with the evidence attached, and the team that owns the resource decides.

claude · checkout-api

>

Which databases in checkout-api are oversized for what they actually use?

●

costory - get_context (MCP)()

⎿

… +24 lines (ctrl+o to expand)

●

costory - query (MCP)(queries: [{type: "cost"}], filterCel: "cos_feature == 'checkout-api'", groupBy: "cos_service_name")

⎿

… +16 lines (ctrl+o to expand)

●

Rightsize the checkout primary from db.r6g.4xlarge to db.r6g.2xlarge.

- CPU p95 is 22% and memory is 41% over the last 30 days.

- Estimated impact: about €1,100/month at current usage.

What is a FinOps MCP?

Cost context your assistant can actually use

The Model Context Protocol (MCP) lets AI assistants call external tools. A FinOps MCP exposes cost data and cost actions. Provider MCPs return raw, provider-native billing data. Costory returns normalized spend across providers, already allocated to your teams and products, next to the deploys and events that explain it.

# Claude Code

claude mcp add --transport http costory \

https://app-api.costory.io/mcp

# then run /mcp and choose Authenticate

Claude Desktop and claude.ai: Settings → Connectors → Add custom connector, then paste the URL. Login is OAuth with your Costory account.

Provider MCP vs Costory

Raw billing data or allocated, explained cost?

Raw provider billing MCP

Costory FinOps MCP

Data shape

Provider-native line items

Normalized, allocated spend across AWS, GCP, Azure, Kubernetes, data platforms and LLM APIs

Scope filters

Account, service, tag (if tagged)

Team, product, environment via Virtual Dimensions

Explain cost changes

Billing deltas only

Cost diffs plus deploys, metrics and custom events

Actions from chat

Query only

Alerts, reports, dashboards, event annotations

Setup

IAM roles, local server or Docker

One URL and OAuth

What the provider MCPs cover today

AWS

The AWS Billing and Cost Management MCP server runs locally with your AWS credentials and covers Cost Explorer, Budgets, anomaly detection and optimisation data for AWS only.

Azure

Microsoft’s Azure MCP Server includes an optional Cost Management toolset for Azure spend.

Google Cloud

Google’s Cloud Billing MCP covers billing accounts, IAM and the pricing catalog. It does not return spend data.

Based on each provider’s public documentation, checked October 2026. Provider MCPs are free and a good fit for single-cloud teams that only need raw billing data; you can run them alongside Costory.

MCP tools

The tools the agent works with

A selection of the tools the Costory MCP server exposes. The full reference is in the docs.

A selection of the tools the Costory MCP server exposes. The full reference is in the docs.

Costory’s FinOps skills are included automatically through the MCP, nothing to install.

01

get_context

Workspace overview: available dimensions, popular group-bys and recent dashboards. Called first in each conversation.

02

query

Cost, business metrics, usage, budgets, period comparisons and formulas in one query.

03

search

Full-text search across dimension values, events, alerts, dashboards, budgets and metrics.

04

list_events

Deploys, provider events and custom events, to line up changes with cost.

05

create_alert

Create cost alerts from chat; preview_alert checks them before saving.

06

create_report / create_dashboard

Save recurring reports and dashboards that the team can open in Costory.

Full MCP tools reference

Works with

Runs in the assistant your team already uses

Documented setups for Claude, Claude Code, Cursor, GitHub Copilot in VS Code and Dust. Grok Bot connects through its MCP connectors, and any MCP agent that supports remote MCP servers with OAuth can connect.

Prompt

Why did our AWS cost go up last week? Break it down by team and service, and list deploys in the same window.

Prompt

What is our cost per customer this month, and which product line drives the change?

Prompt

Alert #finops if Amazon EC2 rises more than 10% versus the trailing 7 days.

Once set up, Costory’s agent runs its playbooks on its own. These prompts are for digging in yourself; they run against your own data once Costory is connected.

FAQ

FinOps MCP questions

Does the MCP get write access to our cloud?

No. Costory reads billing exports. MCP actions create Costory objects only: alerts, reports, dashboards and events.

How does authentication work?

The server is hosted at app-api.costory.io/mcp and uses OAuth with your Costory account. Answers are scoped to your workspace role and team.

Do I need a Costory account to use the Costory MCP?

Yes. The Costory MCP is hosted and works with your Costory account: it serves your organization’s allocated cost data and Context layer. Start with the 30-day free trial or the setup prompt.

Where is the data stored?

Costory stores and processes data in the EU.

Can we use it with the provider MCPs?

Yes. Some teams keep the AWS MCP for raw AWS lookups and use Costory for allocated, multi-provider questions.

Why not connect the AWS MCP to Claude myself?

You can, and for one person on one AWS account it works. It stops working with more than one user (everyone rebuilds prompts, credentials and definitions), with a multi-provider stack (cloud, LLM and data cloud need one normalized, allocated model), and with several teams or features (allocation rules change with reorganizations and M&A). Costory covers those three from Starter, €250/month.

Which assistants work with Costory?

Claude, Claude Code, Cursor, VS Code with GitHub Copilot and Dust connect to the hosted Costory MCP server with OAuth. Grok Bot connects through its MCP connectors, and any other MCP agent can use the same remote URL. Slack works without any assistant setup.

Connect in under 30 minutes. Get your first optimization recommendation the next day.

30-day free trial. Plans from €250/month, billed annually. We recommend Team (€600/month) once more than one team uses Costory.

First optimization recommendation

The next day

ROI

Pays for itself with the first recommendation you act on

Setup

About 30 minutes

Data residency

EU

Access

Read-only on your cloud