Native cloud MCPs
One MCP per provider
Read-only queries, no workflow
Allocation rebuilt per query
No event correlation, no "why"
Verbose metadata, token-heavy
Costory
One endpoint across AWS, GCP, Azure, Datadog, OpenAI, Anthropic, Cursor
Query, alert, annotate, send to Slack
Persistent team and product allocation
Deploys, pricing changes, traffic attached at ingest
Compact LLM format, pre-compiled scopes
Honest take
AI assistants are a step forward, but still not enough
Claude Code with a native cloud MCP is genuinely good at "what changed." For one provider, one team, one person asking, it does the job. But it doesn't scale.
The breaking points
Where the DIY setup breaks
1
No persistent allocation. Every query rebuilds "who owns what" from scratch.
2
No cross-provider normalization. AWS CUR, GCP billing, Datadog, LLM invoices: 4 taxonomies, no unified query.
3
No "why". Raw numbers, no deploy or pricing context. The manual work you wanted to automate.
4
No historical baselines. Every spike looks the same. No "is this normal."
5
Single point of failure. The builder is the setup. PTO, churn, reorg, and it's gone.
6
Agent-hostile at scale. 50 agents reconstruct allocation 50 times, 50 different answers.
7
Hidden TCO. 2-5 days to build, 2-4 days per quarter to maintain. €15K-€30K/year in eng time.
Head-to-head
MCP head-to-head
Native cloud MCPs | Costory MCP | |
|---|---|---|
Coverage | One MCP per provider | AWS, GCP, Azure, Datadog, OpenAI, Anthropic, Cursor |
Context | Raw numbers | Event correlation, business metrics |
Allocation | None | By team, product, business unit and more |
Workflow | Read-only | Query, alert, annotate, push to Slack, BI, IDP |
LLM efficiency | Verbose metadata | Compact format |
Total cost of ownership: DIY vs Costory
DIY | Costory | |
|---|---|---|
Initial build | 2 to 5 days | 30 minutes |
Quarterly upkeep | 2 to 4 days | Barely any |
Annual cost | €15,000 to €30,000 (Eng salary) | €3,000 flat |
Second user | Context in builder's head | Shared endpoint |
Fleet of agents | Each rebuilds logic | Same endpoint |
Scale
What happens as agent count grows
Cursor 3 already runs parallel cloud agents — 35% of their PRs are agent-generated. The ratio of agents to humans querying cost data is going up.
DIY + native MCPs
1 agent: works — one query, one prompt, one builder.
10 agents: allocation re-derived per prompt, answers diverge. Cold cache, fragmented audit, 10 prompts to patch on schema change.
50 agents: 50 identities to govern, no central audit, schema-change blast radius ×50. Annual rebuild cost > €3K/year.
Costory
1 agent: works. Same endpoint your humans use.
10 agents: one context layer, shared cache, single audit trail.
50 agents: flat governance surface, one source of truth. Total cost = €3K/year.
Fair use
When DIY is the right call
One cloud provider
Cost Explorer plus a native MCP covers 80% of a single-provider question set.
One person asking
The "context in your head" problem doesn't exist until someone else needs the answer.
No LLM spend worth allocating
If OpenAI, Anthropic, and Cursor aren't line items yet, the cross-provider gap hasn't opened.
Setup
Start using Costory in 30 minutes with Terraform
One terraform apply to connect every provider. Allocation runs on the tags you already maintain.
Features
What replaces the workaround
MCP server
Structured tools across every provider, queryable from Claude, Cursor, or any MCP client.
Virtual dimensions
Allocation logic that layers on top of your existing tags, without a re-tagging project.
Event correlation
Deploys, pricing changes, and traffic spikes tied to cost movements at ingest.
Slack reports
Weekly per-team digests and drift alerts, delivered in channel.
Cost Explorer
Drill into any line, share a saved view, jump to the deploy that caused the spike.
BI exports
Native PowerBI, Looker, BigQuery. Your existing reporting keeps working.
Pay for the context layer, not the dashboards
€250 / month flat — up to €10M annual spend
No percentage of cloud spend
No enterprise sales call
No need to hire a FinOps team
FAQ