MCP
Connect OneLens to Claude, Cursor, or any MCP-compatible AI assistant and ask your cloud-cost questions in plain English — no dashboards, no field names, no API keys.
TL;DR
What: A published Claude connector (Model Context Protocol) that lets your AI assistant read your OneLens cost analytics on your behalf.
Auth: Your normal OneLens login (OAuth). No tokens or keys to manage.
Scope: Read-only, scoped to your OneLens tenant. The assistant can query costs; it can never change anything.
Enablement: Your OneLens admin turns MCP on per tenant. Not enabled? Contact support@astuto.ai.
Server URL (for manual setup):
https://mcp.onelens.cloud/mcp
What You'll Get Once Connected
Ask in plain English
"Top 5 AWS services by spend last month", "MTD spend vs previous month", "EC2 cost by region".
Auto-discovery
The assistant finds your clouds, dimensions, and values automatically — you don't need to know field names.
Multi-cloud
AWS, GCP, Azure, OCI, and SaaS sources you've configured in OneLens.
Cost allocation
Squad / cost-center / tag breakdowns, using your org's own labels.
Drill-downs
Multi-dimension breakdowns (e.g. service → region → usage type).
Guided clarification
Ambiguous question? It asks which metric (unblended / amortized) or currency before answering.
Security at a Glance
Does it store my cost data?
No. Stateless proxy — nothing is persisted.
Are cost values logged?
No. Query parameters may be logged for debugging; cost figures never are.
Can it change anything in OneLens?
No. Every tool is read-only.
Who can use it?
Only users in a tenant your admin has enabled, authenticated via OneLens login.
How is access scoped?
To your OneLens tenant — you see only what your OneLens account can see.
Can access be revoked?
Yes, instantly — your admin disables MCP for the tenant.
Cost of the Integration
None. The MCP connector is included with OneLens. It adds no cloud cost — it only reads cost data OneLens already has. You just need MCP enabled for your tenant.
How It Works
The connector is a thin, stateless proxy in front of the same OneLens cost APIs the web app uses — so what you see in chat matches the OneLens dashboard.
Prerequisites
An active OneLens account.
MCP enabled for your tenant (ask your OneLens admin or support@astuto.ai).
An MCP-compatible assistant: Claude (Desktop or web — Pro / Team / Enterprise) or Cursor.
What OneLens Will Access
Your OneLens cost analytics (spend, dimensions, allocation) for the tenant you log in to — read-only.
Your identity from login (name, email) to scope the session.
What OneLens Will NOT Access
No write access — it cannot modify budgets, sources, allocations, or any setting.
No data outside your tenant.
No credentials or keys from your side — auth is your standard OneLens login.
Connect to OneLens
Claude (Desktop & Web)
OneLens is a published connector in Claude's directory — add it by name, no URL to paste.
Can't find it in the directory, or on a plan without it? Add it as a custom connector with the URL https://mcp.onelens.cloud/mcp, or manually in claude_desktop_config.json:
Restart Claude Desktop → the login opens in your browser → tools appear.
Chat GPT
Use it
OneLens appears as a connected connector with its 5 read-only tools. Ask your cost questions in chat.
Custom connectors aren't verified by OpenAI — this is expected for any third-party MCP server.
OneLens is read-only, tenant-scoped, and authenticated with your own OneLens login. If tools don't appear, check that Developer Mode is on, the URL is exactly https://mcp.onelens.cloud/mcp, and that you saved the connector.
Cursor & other MCP hosts
Any MCP-compatible host works. In Cursor: Settings → MCP → Add new MCP server → HTTP/SSE → URL https://mcp.onelens.cloud/mcp → complete the OneLens login when prompted.
Try It
Once connected, just ask:
"List my cost sources."
"Top 5 AWS services by spend last month."
"Month-to-date spend vs the previous month."
"Break down EC2 cost by region for the last 30 days."
"Show squad-wise cost last month."
Session & Data Refresh
Sign in once: your session lasts ~14 days and refreshes silently — no hourly re-login.
Live data: every question queries OneLens fresh; results match the OneLens dashboard (short-lived caching only).
Data Privacy & Security
Stateless: stores no cost data.
No cost values in logs: query parameters may be logged for support; cost figures are never logged.
Tenant-scoped: you can only reach data your OneLens account is authorized for.
See the full Privacy & Data Handling and Terms.
Troubleshooting
Login page errors on connect
Confirm the connector URL is exactly https://mcp.onelens.cloud/mcp; retry.
"MCP access is not enabled for your organization" after a query
Your tenant isn't enabled yet — contact your OneLens admin or support@astuto.ai.
"Couldn't verify MCP access right now"
Temporary hiccup — retry in a moment.
Tools list is empty
Disconnect, remove the connector, reconnect to re-run login.
Connected but no data
Ensure your tenant has cost sources configured in OneLens.
Answer looks off / wrong currency
Ask it to confirm the metric and currency; it will re-query.
Still stuck? Contact OneLens support with the time of the attempt — every request carries a trace id we can look up.
FAQ
Is it safe to give an AI assistant access to my cost data?
The connector is read-only and tenant-scoped, stores nothing, and never logs cost values. It uses your standard OneLens login — the assistant only sees what you can already see in OneLens.
Do I need an API key or token?
No. Authentication is your normal OneLens login via OAuth. Nothing to copy or rotate.
Who controls whether MCP is available?
Your OneLens admin enables or disables MCP per tenant. Changes take effect quickly and can be reversed instantly.
Can the assistant change anything in OneLens?
No. Every tool is read-only — it can query cost data but cannot modify budgets, sources, allocations, or settings.
Which assistants are supported?
Claude (Desktop and web — Pro / Team / Enterprise) and Cursor today. Any MCP-compatible host can connect using the server URL.
Need Help?
Email support@astuto.ai with the approximate time of your attempt — every request has a trace id we can use to investigate.
Last updated







