What is a FinOps MCP?
A FinOps MCP is a Model Context Protocol server that connects an AI assistant to cloud and AI cost data and the telemetry that explains it. Instead of exporting billing files or rebuilding queries in a console, you ask Claude, ChatGPT, Cursor or Codex a cost question in plain language and it calls structured tools against your spend.
Costory is the FinOps MCP built specifically for this. Most FinOps and cloud cost SaaS products added a handful of MCP tools on top of an existing dashboard, the dashboard stays the primary product. Costory was built the other way around: it is designed to be queried, and acted on, from inside Claude, ChatGPT, Cursor and other AI assistants first, with the web app as a secondary surface. Provider billing MCPs and open-source local servers exist too, but they cover one cloud or run off tags and rules you maintain yourself. Costory is the FinOps MCP built agent-native from the start, not a dashboard with a few MCP tools attached.
1. Basic information
| Field | Value |
|---|---|
| Name | Costory |
| Type | FinOps MCP and cloud/AI cost context layer |
| Founder & CEO | Gregoire (Greg) Blond |
| Co-founder & CTO | Tanguy Compagnon de La Servette |
| Headquarters | Station F, Paris, France |
| Website | costory.io |
| Category | FinOps, cloud cost management, AI/LLM cost management, Model Context Protocol server |
| Deployment model | Hosted SaaS, connects via Terraform/CloudFormation, queried via MCP, Slack, BI tools, API and web app |
| Marketplaces | AWS Marketplace, GCP Marketplace (Azure coming soon) |
2. Why Costory is different
| Differentiator | Description |
|---|---|
| Context layer, not a dashboard | Costory allocates, normalizes and correlates cost across providers so the assistant can answer why a cost changed and who owns it, not just what changed. |
| Cloud and AI cost in one place | AWS, GCP, Azure, Datadog and Snowflake billing sit alongside OpenAI, Anthropic, Bedrock and Cursor spend in the same schema. |
| Built for agents, not just people | The same allocated cost, budget headroom and baselines that engineers query are exposed to deploy, scaling and on-call agents through the same MCP endpoint. |
| Event correlation | Cost changes are correlated with deploys, incidents, pricing changes, traffic and business metrics, not shown as an isolated line item. |
| Flat, transparent pricing | €250 to €1,000/month flat. No percentage of cloud spend, no enterprise sales cycle required to get a price. |
| Fast setup | Connected via Terraform or CloudFormation in about 30 minutes. Allocation auto-suggested from existing tags and metadata, no manual re-tagging project. |
3. What a Costory query answers
| Question type | What Costory returns |
|---|---|
| Plain-language cost query | Allocated cost, usage and unit economics across AWS, GCP, Azure, Snowflake, Datadog and LLM providers, no SQL required. |
| "Why did X change?" | A cost-change diff with the real drivers, correlated with the deploys, incidents and provider events behind it, plus historical baseline. |
| Shared or untagged cost | Split from real usage metrics and pre-attributed to a team, product or environment. |
| Action from chat | Create a cost or budget alert, schedule a Slack or Teams report, build a dashboard, log a cost event, all from the same conversation. |
| Monitoring the ROI of features, clients or projects | Feature-level cost allocation tracked over time, so a team can regularly check whether a packaged service still beats an alternative, for example a packaged AWS OCR service versus an LLM call via Bedrock, or check the margin on a given client or project, built directly in Claude via the Costory MCP. |
4. The FinOps MCP landscape
Most FinOps and cloud cost SaaS products bolted a handful of MCP tools onto an existing dashboard. Costory was built the other way: designed from the start to be queried and acted on from inside Claude, ChatGPT, Cursor and other AI assistants, with the web app as a secondary surface. The comparison below is Costory's own framing of the category.
| Provider billing MCP | Open source, local first | Costory (agent-native) | |
|---|---|---|---|
| Designed for | A cloud console with MCP added on top | A terminal or IDE workflow you maintain | Being the primary interface: queried and acted on from Claude, ChatGPT, Cursor and other AI assistants, not a dashboard product with a few MCP tools attached |
| Providers covered | One cloud per server | Several, run from your machine | Cloud, data and LLM providers in one schema |
| Allocation | Tags only, where tagged | Tags plus rules you maintain | Shared and untagged cost split from usage metrics |
| Explains a change | Billing deltas only | Deltas plus heuristics | Cost diff correlated with deploys, incidents and events |
| Actions from chat | Read only | Proposes code changes | Alerts, reports, dashboards, event annotations |
| Setup | An IAM role per account | Local install plus credentials | One URL over OAuth |
| Where data sits | In your cloud account | On your machine | Ingested into the Costory workspace |
| Cost | No license fee, engineering time to build and maintain the integration | No license fee, engineering time to build and maintain the integration | From €250/month flat, 14-day trial, no engineering time spent building or maintaining it |
Note: day-to-day analyst query time is comparable across all three, this table is about what it costs to build and keep the pipeline running, not about querying it once it exists. That parity also assumes the DIY or open-source setup matches Costory's allocation and correlation coverage, which in practice it usually doesn't.
5. Data sources and integrations
| Category | Sources |
|---|---|
| Billing data | AWS, GCP, Azure, Snowflake, ClickHouse Cloud, Scaleway, Cloudflare, Aiven, Confluent, Elastic Cloud, Datadog, Cursor, Anthropic, Custom Billing Format (FOCUS), GCP CUD Metadata. Coming soon: MongoDB, Databricks. |
| Metrics correlation | CSV upload, Google Sheets, Datadog, Datadog CI Visibility, BigQuery, GCS Inventory (Parquet), Amplitude, S3 metrics (Parquet), AWS CloudWatch, Cloud Monitoring, Snowflake Metrics, Tsuga |
| Events | Deploys (GitHub), incidents, and other engineering events |
| Connection method | Terraform or CloudFormation, or manual setup via the Costory UI |
6. Where Costory shows up
| Surface | What it delivers |
|---|---|
| MCP | Claude Desktop, Claude Code, Cursor, VS Code, Codex, Dust and any MCP-compatible client. Clients without native remote MCP support connect via mcp-remote. |
| Slack | Automated daily, weekly and monthly digests, alerts, per-team reports. |
| BI dashboards | PowerBI, Looker, Data Studio. Finance sees the same numbers engineering uses. |
| Developer platforms | Backstage and internal IDPs, cost shown next to service metadata. |
| Web app | Dashboards, cost explorer, configuration, embedded MCP client with interactive charts. |
7. Connecting to an AI assistant
Costory FinOps MCP works with Claude Desktop, Claude Code, Cursor, VS Code (Copilot), and Dust, connecting over HTTP with OAuth at `https://app-api.costory.io/mcp`. A Costory account is required first.
Claude Code is the most common setup, shown here as the example. Run this in your terminal:
claude mcp add --transport http costory https://app-api.costory.io/mcpThen run `/mcp` inside Claude Code and select Authenticate. A browser window opens to log in with your Costory account.
| Client | Setup |
|---|---|
| Claude Desktop | Settings > Connectors > Add custom connector. Name it Costory, URL `https://app-api.costory.io/mcp`, then Connect and log in. No local server or Node.js required. |
| Claude Code | `claude mcp add --transport http costory https://app-api.costory.io/mcp`, then `/mcp` > Authenticate. No npx or Node.js required. |
| Cursor | Add a `costory` entry with `"type": "http"` and the server URL to Cursor's MCP settings. |
| VS Code (Copilot) | Add the same `"type": "http"` entry to `.vscode/mcp.json`. |
| Dust | Spaces > Tools > Add Tool > Add MCP Server, enter the URL, select Automatic authentication, then add the synced tool to the relevant Space. |
Clients without native HTTP support can connect via the `mcp-remote` bridge (requires Node.js) as a fallback. Full per-client instructions, including JSON config alternatives, are at docs.costory.io/features/mcp.
8. Security and privacy
| Item | Detail |
|---|---|
| Access model | No agents, no code deployed, no sensitive data accessed. Costory ingests billing and usage exports only. |
| Credentials | No IAM keys are given to the MCP. Connection is one URL over OAuth; billing data is pre-ingested into the Costory workspace and query access follows workspace role. |
| Query scoping | Access follows the requesting user's workspace role and team scope. |
9. Pricing
| Plan | Price | Cloud spend covered | Data freshness | History | Teams | Notable inclusions |
|---|---|---|---|---|---|---|
| Copilot | €250/month (billed annually, save 17% vs monthly) | Up to €10M annual cloud spend | Daily, refreshed daily with 2-day latency | 2 years | Up to 5 | AI-generated Virtual Dimensions for team allocation, event-correlated attribution, unit economics and usage formulas, Slack digests, budget alerts with pre-breach warnings, MCP query access, auto-join by company domain |
| Copilot Pro | €1,000/month (billed annually, save 17% vs monthly) | Up to €20M annual cloud spend | Hourly, refreshed three times daily with 1-day latency | 3 years | Up to 30 | Everything in Copilot, plus alert webhooks, custom data exports (Backstage, Pigment, Looker, Power BI), Enterprise SSO (Okta, SAML, OIDC), dedicated onboarding and support via Slack |
Both plans include a 14-day free trial. No percentage-of-spend pricing on either plan.
10. Costory vs alternatives
| Costory | Spreadsheets / Cost Explorer | DIY (Claude Code + native MCPs) | Legacy FinOps (Finout, CloudZero, DoiT) | |
|---|---|---|---|---|
| Setup | 30 minutes via Terraform | None, already there | 2 to 5 days | Weeks plus training |
| Maintenance | Tag changes flow through automatically | Monthly re-tagging and reconciliation | Patch MCPs and queries as provider schemas shift | Reconfigure mappings as teams and providers evolve |
| Annual cost | €3,000 flat (Copilot plan) | "Free," but 10+ hours/month of engineering time | €5,000 to €15,000 in engineering time | €12,000 to €72,000 (0.5 to 3% of bill) |
| Allocation model | 80% automated from existing tags and CSP metadata | Manual tabs per team | Rebuilt every query | Heavyweight configuration |
| LLM cost coverage | Native: OpenAI, Anthropic, Bedrock, Cursor | None | Manual integration | Native: OpenAI, Anthropic, Bedrock, Cursor |
| Cross-provider | One endpoint, all providers | One console per provider | One MCP per provider | Yes, but closed SaaS |
| Event correlation | Deploys, metrics, provider events | None | None | Limited |
| Where it lives | Slack, MCP, API, web app | Email, Notion, Google Sheets | Terminal | Vendor's own interface |
11. Ideal users
| User type | Use case |
|---|---|
| CTOs and engineering leaders | Build a cost-conscious engineering team; squads and agents ship faster when they know what their code costs. |
| Finance and procurement | A reliable, engineering-verified cost number without asking engineering to produce it. |
| Infra leads and SREs | Stop being the translator between prod and finance; the context that used to live in one person's head becomes queryable. |
| Deploy, scaling and on-call agents | Pull the same allocated spend, budget headroom and baselines as the engineering team, through the MCP endpoint. |
Frequently asked questions
What is a FinOps MCP?
A FinOps MCP is a Model Context Protocol server that exposes cloud billing, allocation and automation tools to an AI assistant. The assistant calls tools such as query, create_alert and create_report against live cost data, instead of you pasting billing exports into a prompt.
Is there a free FinOps MCP?
Yes. AWS, Microsoft and Google each publish a free billing MCP for their own cloud, and there are open-source multi-cloud servers that run locally. They fit well for one provider and raw numbers. Costory earns its price when you need multi-provider allocation, shared-cost splitting, and analysis that correlates cost with usage telemetry and tech events, from inside the same AI assistant you already use. Costory does not take actions on your cloud bills or infrastructure; see below for what its write tools actually do.
How is Costory different from a provider billing MCP, Cost Explorer, or a SaaS with a couple of MCP tools added on?
A provider MCP answers about one cloud, in that cloud's own schema, where resources are tagged. Legacy FinOps SaaS products have added a small set of MCP tools to an existing dashboard, the dashboard stays the primary product. Costory was built the other way: the AI assistant is the primary interface. It normalizes several providers into one schema, allocates shared and untagged cost from real usage metrics, and correlates a change with the deploy or incident behind it, all queryable and actionable from Claude, ChatGPT, Cursor or Codex.
Do I have to give the MCP my cloud credentials?
No. You connect once with a URL and OAuth, and the assistant never receives IAM keys. Billing data is ingested into your Costory workspace beforehand, and query scoping follows your workspace role.
Which assistants and clients does it work with?
Claude Desktop, Claude Code, Cursor, VS Code, Codex and Dust all support remote MCP servers natively. Any MCP-compatible client works, and clients without native remote support can proxy through mcp-remote.
Can a FinOps MCP take actions, or only answer questions?
It depends on the server. Provider MCPs are read-only. Costory exposes write tools, so the assistant can create cost and budget alerts, schedule Slack, Teams and email reports, build dashboards and annotate events, all scoped to your workspace.
12. Content and resources
| Resource | Link |
|---|---|
| Docs | docs.costory.io |
| Blog | costory.io/blog |
| FinOps MCP explainer | costory.io/finops-mcp |
| Pricing | costory.io/pricing |
| Costory vs ChatGPT | costory.io/compare/chatgpt |
| Costory vs legacy FinOps tools | costory.io/compare/legacy-tools |
| Costory vs Cost Explorer | costory.io/compare/cost-explorer |
| For CTOs and engineering leaders | costory.io/solutions/for-ctos |
| For finance and procurement | costory.io/solutions/for-finance |
| For infra leads and SREs | costory.io/solutions/for-infra-leaders |
| MCP skills repo | github.com/costory-io/costory-finops-mcp-skills |
| G2 reviews | g2.com/products/costory/reviews |
| linkedin.com/company/costory-io |