Clara MCP - nOps Documentation

Overview

Clara is nOps’s FinOps assistant. In the product you can chat with Clara on the Ask Clara page, from the Ask Clara button on several pages, and on canvases. The Clara MCP brings that same assistant into the AI tool you already work in: Cursor, Claude, or any other client that supports the Model Context Protocol (MCP). Add one address, sign in once, and you can:

Ask about spend, savings, usage, and commitments in plain language
Work with your reports, dashboards, budgets, cost allocations, unit economics, contracts, and MAP projects without leaving your editor
See anomalies and optimization recommendations as soon as nOps finds them
Put an analysis on a schedule so the answer shows up in Slack or email on its own

You need an active nOps account and the same data access you have in the web app. MCP does not bypass organization permissions.

Connect in 1 minute

Clara MCP is one address. Add it to your AI app, sign in with your nOps account, and start asking.

Connector address

https://mcp.nops.io

There is no API key. The first time you connect, your AI app opens the nOps sign-in page (signin.nops.io). Sign in, and you’re done.

Ask your AI agent to set it up

Copy this prompt into your AI app, or send it straight to Claude or ChatGPT:

Prompt

Help me add nOps Clara to this AI app. The steps are on https://help.nops.io/plugins/clara-mcp#connect-in-1-minute. The name is nOps Clara. The URL is https://mcp.nops.io. It uses sign-in with my nOps account. If you can use my web browser, do the steps for me. If you cannot, tell me what to click.

Send to Claude

Opens a new Claude chat with the prompt filled in.

Send to ChatGPT

Opens a new ChatGPT chat with the prompt filled in.

Or follow the steps for your app below.

Claude (claude.ai and Claude Desktop)

Open Connectors

Go to Customize → Connectors → Add custom connector.

Add Clara

Name it nOps Clara and paste https://mcp.nops.io as the URL. Click Add.

Connect

Click Connect and sign in to nOps when prompted.

Claude Code (CLI): run this once, then sign in when Claude Code asks.

claude mcp add --transport http clara https://mcp.nops.io

ChatGPT

Open Plugins

Go to chatgpt.com/plugins and click Add, then Add custom MCP server.

Add Clara

Name it nOps Clara, paste https://mcp.nops.io as the Server URL, and set Authentication to OAuth.

Create and connect

Tick the acknowledgement, click Create, then Connect and sign in to nOps.

Custom MCP servers depend on your ChatGPT plan and workspace settings. Free and Plus plans may not offer them. If you don’t see Add custom MCP server, ask your ChatGPT workspace admin.

Cursor

Add Clara to Cursor opens Cursor and installs the server in one click. Then ask Clara anything and sign in to nOps when prompted. You can also install it from Cursor Settings → Plugins, or add it by hand (see Other MCP clients).

Start a chat

Once you’re connected, try one of these:

Ask	Claude	ChatGPT
”List my Clara datasets.”	Open	Open
”Break down our AWS spend last month by service and account.”	Open	Open
”How are our Savings Plans and Reserved Instances performing this year?”	Open	Open
”Any unusual spend in the last two weeks, and what drove it?”	Open	Open
”What is nOps recommending we buy, and what would it save?”	Open	Open

More ideas are in Things to try.

Good to know

Sign in to the nOps web app at least once first. Clara MCP uses the same account.
Your organization needs Clara MCP turned on. If sign-in works but Clara refuses to connect, ask an org admin to enable it.
Add connectors on the web or desktop app. Once added, they also work in the mobile apps.
Your nOps role applies. You see only the data you can see in the web app, and changes need Admin or Owner (see What you can change).

Capabilities

Clara MCP works with your real nOps data and the things you’ve built on top of it: reports, budgets, cost allocations, contracts, and more. Here’s what you can do, grouped by the kind of work. Some actions (create, update, delete, send) change things in your nOps account and need an Admin or Owner role. Those are bolded below, and you can read more in Read vs. write access.

Asking questions about your spend

Clara answers from a set of prepared views of your cost data, called datasets. Each one covers a topic (AWS billing, commitments, AI tool spend) with a known set of fields to group and filter by, so the same question always comes back the same way.

You can	What happens
See which datasets you have	Clara lists what your account can query.
Look inside a dataset	Clara shows the fields and metrics available, so you know what you can ask for.
Ask a question	Clara runs it and returns the rows. Your data stays scoped to your organization automatically.

A good first message after connecting: “List my Clara datasets.” Then “What’s in the Commitment Management Analysis dataset?” before you dig into spend or savings.

What you’ll see depends on your setup: which clouds you’ve connected, whether Commitment Management and ShareSave are turned on, which third-party tools you’ve integrated, and your role. Typical datasets:

Dataset	What you can ask about
Explorer Cost & Usage	Your detailed AWS bill: spend and usage over time, by account, service, tag, purchase type, and more. The best starting point for most questions.
Commitment Management Analysis	How your Reserved Instances and Savings Plans are performing: savings rate, coverage, savings vs. on-demand pricing, and how much commitment is going unused.
Commitment Burndown	What AWS commitments you still hold, how well they’re being used, and when they run out.
GCP Commitment Inventory	The same view for your GCP committed use discounts.
Recommendations	What nOps suggests you buy for Compute, Databases, and AI, and how much it would save per year.
Savings Realizations	Savings you’ve actually realized through ShareSave Risk-Free (excludes the current month).
Real-time Billing Commitment Management	The same savings picture with current-month usage included, up to the last few days.
AWS Cost Anomalies	Unusual spend AWS has flagged: how much extra it cost, how serious it is, and the account, region, and usage type behind it.
Cursor Cost & Usage	Cursor spend and usage by person, model, and hour. Useful for charging teams back and spotting unused seats. Needs the Cursor integration.
Claude.ai Enterprise Cost & Usage	Claude.ai Enterprise spend and usage by person, model, and hour, including each person’s spend limit. Needs the Claude Enterprise integration.

Exploring spend and building reports

The everyday work you’d do in Inform: dig into costs, save a view, share it with your team, without switching windows.

You can	In nOps	What happens
Explore costs	Explorer	Totals, trends over time, and heatmaps of your spend and usage, plus a list of the values you can filter on.
Compare two periods	Explorer	Clara shows what changed between the two, and what drove the change.
Work with dashboards	Explorer Dashboard	See your dashboards, and create, update, copy, or delete them.
Work with reports	Reports	See your saved reports and their numbers, create, update, or delete them, and send one to your team.
Check recommendations and anomalies	Recommendations, Anomalies	See the savings opportunities and spend anomalies nOps has found.
Manage MAP projects	MAP Manager	See eligible accounts and project resources, and create, update, or delete projects.

Splitting costs across teams

Assign spend to the teams, products, or departments that own it, so everyone sees their share.

You can	What happens
Manage cost allocations	See your allocations and totals, and create, update, copy, or delete them.
Manage the rules	See how spend is assigned, and create, update, reorder, or delete rules.
Manage the buckets	Change the set of teams, products, or departments an allocation splits across.
See the results	Get the dollar amount per bucket, the resources inside a bucket, and how the split has changed over time.
Preview before you commit	Try a rule change and see the resulting split before saving it.
Keep up with new values	Find newly available values (a new tag or account) and pull them into your existing rules.

Budgets, unit economics, and contracts

The guardrails you set in Operate: what you plan to spend, what a unit of your business costs, and what you’ve committed to your cloud provider.

You can	In nOps	What happens
Manage budgets	Budgets	See budgets and how they’re tracking, create, update, or delete them, export to CSV, and send one out.
Set up alerts	Budgets	Create, update, or delete the thresholds that notify you, for example at 90% of budget.
Plan next period	Budgets	Build a plan line by line, or have nOps generate one from your spend history.
Track unit economics	Business Unit Economics	See cost per unit of your business, create or update the setup and its numbers, preview spend, and send results.
Track contracts	Contract Tracker	See your cloud contracts and how much you’ve burned through, and create, update, or delete them.

Commitments and savings

How your Reserved Instances and Savings Plans are doing, and what nOps thinks you should buy next: the same picture as Optimize.

You can ask about	What you’ll get
What you hold	The commitments still on the books, how well they’re used, and when they expire.
How they’re performing	Your savings rate, how much of your usage is covered, and the savings you’ve actually realized.
What to buy next	nOps purchase recommendations and the savings each one projects.

AI and LLM spend

What your AI tools cost, and whether the seats you’re paying for are being used.

You can	What happens
Spot AI cost spikes	See anomalies specific to your AI and LLM spend.
Set your AI view	See and change the dashboard your team lands on for AI costs.
Track seats	See who has a seat for each AI tool and how much they’re actually using it.

Answers on a schedule

Set a question up once and have the answer show up on its own: in Slack or Teams, by email, or in your Ask Clara chats.

You can	What happens
Schedule a question	Save a recurring question to Clara, turn it on or off, run it now, or delete it.
Schedule a delivery	Have a report or budget sent out on a cadence to Slack, Teams, or email.
Check what ran	Review your past runs and the result of any single one.

Things to try

If you want to	Ask
Get oriented	”List my Clara datasets.”
Understand last month’s bill	”Break down our AWS spend last month by service and account.”
Check commitment performance	”How are our Savings Plans and Reserved Instances performing this year?”
Find coverage gaps	”Where is our commitment coverage below 90%?”
See what’s expiring	”How much commitment value do we still hold, and when does it run out?”
Find more savings	”What is nOps recommending we buy, and what would it save?”
Investigate a spike	”Any unusual spend in the last two weeks, and what drove it?”
See who owns what	”How is our spend split by team, and what’s unallocated?”
Put a limit in place	”Create a budget for our Cursor spend and alert me at 90%.”
Review AI tool costs	”Who are our top ten Cursor users by cost this billing cycle?”
Stop asking manually	”Send me this commitment coverage report in Slack every Monday morning.”

Plugins with skills (optional)

The address above is all you need to use Clara. The plugin zips add skills and rules on top: instructions that teach Cursor or Claude how to use Clara’s tools well, such as running a full commitment analysis. Install a plugin if you want those extras.

Where to get the plugins

Sign in to nOps and open Settings.
Under Organization Settings, open Plugins, then find Clara MCP & Connectors.
Under Latest release, expand Download assets and grab the zip for your tool (when available):
- clara-cursor-plugin-*.zip: Cursor
- clara-claude-plugin-*.zip: Claude Desktop or Claude Code

You can also download releases from the open-source repo: nops-io/clara-mcp.

Cursor plugin

Steps

Download the Cursor plugin

Get the latest clara-cursor-plugin-v*.zip from Settings → Plugins → Clara MCP & Connectors or GitHub Releases.

Extract into your project

Unzip into your project root (or your home directory for a global install):

unzip clara-cursor-plugin-v*.zip -d /path/to/your/project

This adds .cursor/mcp.json, .cursor/rules/clara.mdc, and .cursor/skills/ where Cursor expects them.

Enable the MCP server

Open Cursor Settings → Tools & MCP and turn on the clara server.

Reload Cursor

Run Cmd+Shift+P → Reload Window (Mac) or the equivalent on Windows / Linux.

Verify

In Cursor Chat, ask: “List my Clara datasets.” On the first request, Cursor prompts you to authorize with nOps (OAuth). Complete login in the browser; later requests reuse your session.

What the Cursor package includes

Path	Purpose
`.cursor/mcp.json`	Registers the Clara MCP server
`.cursor/rules/clara.mdc`	Guidance so Cursor uses Clara tools well
`.cursor/skills/query-clara/`	General FinOps / cost questions
`.cursor/skills/commitment-analysis/`	Savings rate, coverage, RI/SP burndown

Global install (all projects)

Extract to your home directory instead of a single repo:

unzip clara-cursor-plugin-v*.zip -d ~

Claude Desktop plugin

Steps

Download the Claude plugin

Get the latest clara-claude-plugin-v*.zip from Settings → Plugins → Clara MCP & Connectors or GitHub Releases.

Add the plugin in Claude

In Claude, go to Settings → Plugins and click + next to Personal plugins.

Select the zip

Choose the downloaded clara-claude-plugin-v*.zip. Clara appears under Personal plugins in the sidebar.

Connect the MCP connector

Open the Clara plugin → Connectors → Install next to clara, then complete the nOps OAuth prompt.

Verify

Start a conversation and ask: “List my Clara datasets.”

What the Claude package includes

Path	Purpose
`.claude-plugin/plugin.json`	Plugin manifest
`.mcp.json`	Clara MCP server config
`skills/query-clara/`	General Clara queries
`skills/commitment-analysis/`	Commitment performance analysis

Claude Code plugin (CLI)

If you use Claude Code (claude CLI) and want the skills too (for the connector alone, the one-liner above is enough):

Download clara-claude-plugin-v*.zip (same package as desktop).

Install the plugin:

claude plugin install ./clara-claude-plugin-v*.zip

For a single session without a permanent install:

claude --plugin-dir ./clara-claude-plugin-v*.zip

Update when new releases ship:
claude plugin update clara
Connect the clara MCP connector and complete OAuth (same as desktop).

Requirement: Claude Code v2.1.128 or later.

Other MCP clients (manual setup)

For any other client, add an HTTP MCP server pointing at https://mcp.nops.io:

{
  "mcpServers": {
    "clara": {
      "type": "http",
      "url": "https://mcp.nops.io"
    }
  }
}

Client	Typical config location
Cursor	`.cursor/mcp.json` in the project, or `~/.cursor/mcp.json` globally
Claude Code	`.mcp.json` in the project, or `~/.claude/mcp.json` globally
Other	See your client’s MCP documentation

Clara uses OAuth on first use. There is no long-lived API key to paste into the config file.

Authentication

Ask Clara anything in your AI tool (for example “Show my cloud costs for last month”).
On the first request, you’ll be sent to an nOps login page. Sign in with the account that should own the data.
After authorization, the session is stored; you should not need to log in on every message.

Re-authenticate if your session expires or you switch nOps organizations: run a Clara tool again and follow the prompt, or sign out of nOps in the browser and sign back in first. You’ll only ever see the data your nOps role allows. If something is off limits, Clara will say so rather than return it.

What you can change

Clara follows your existing nOps role. There’s nothing extra to set up:

Your nOps role	What you can do
Member	Ask questions and read everything you can see in the app: reports, dashboards, budgets, allocations, contracts, recommendations, and anomalies
Admin / Owner	All of the above, plus creating, updating, and deleting: the bolded actions in Capabilities

If an action fails with a permissions error, check your role with an organization admin, or ask them to make the change for you.

Creating, changing, and deleting through Clara affects your real nOps account, and deletes can’t be undone from your AI tool. It’s worth asking Clara to confirm what it’s about to change before it does.

Clara in nOps vs. Clara MCP

	In nOps (web)	Clara MCP (your AI tool)
Where	Ask Clara, canvases	Cursor, Claude, or another MCP client
Best for	Charts and canvases you share with the team	Quick questions while you work, and anything alongside your code
Setup	None. It’s already there	Add `https://mcp.nops.io`, sign in once

Many teams use both: canvases for dashboards they share, and Clara MCP in the tool they’re in all day.

Troubleshooting

Sign out of nOps in your browser, sign in again, then ask Clara something.
Double-check you signed in to the right nOps organization.

Clara doesn’t show up

Cursor: open Settings → Tools & MCP and make sure clara is switched on, then reload the window.
Claude / ChatGPT: open Connectors (or Plugins in ChatGPT) and make sure nOps Clara is added and connected.
Restart your AI tool after installing a new version of the plugin.

Clara responds but the numbers look empty or wrong

Check that your nOps role can see the data you’re asking about.
For savings questions, end your date range before the current month. Most savings data lags by about a month.
Ask Clara what’s in the dataset. Available fields depend on your integrations and setup.

Network

Clara MCP connects over HTTPS. If you’re behind a corporate proxy or firewall, it needs to allow outbound access to mcp.nops.io and signin.nops.io.

More help

Repo guides: install-cursor, install-claude, auth, troubleshooting
nOps support: support.nops.io or in-app chat

Clara Slack Agent

Ask Clara about spend and commitments in Slack: @mentions, DMs, and scheduled thread delivery.

Ask Clara

Use Clara inside nOps without installing MCP. Ideal when you want charts and canvases in the same session.

Home page

Choose where nOps opens when you sign in.

Integrations

Connect AWS, GCP, and Azure so Clara has cost and commitment data to query.

Plugin releases on GitHub

Download Cursor and Claude plugin zips (and manual MCP config for other clients), release notes, and source docs.

​Overview

​Connect in 1 minute

​Ask your AI agent to set it up

Send to Claude

Send to ChatGPT

​Claude (claude.ai and Claude Desktop)

​ChatGPT

​Cursor

​Start a chat

​Good to know

​Capabilities

​Asking questions about your spend

​Exploring spend and building reports

​Splitting costs across teams

​Budgets, unit economics, and contracts

​Commitments and savings

​AI and LLM spend

​Answers on a schedule

​Things to try

​Plugins with skills (optional)

​Where to get the plugins

​Cursor plugin

​Steps

​Verify

​What the Cursor package includes

​Global install (all projects)

​Claude Desktop plugin

​Steps

​Verify

​What the Claude package includes

​Claude Code plugin (CLI)

​Other MCP clients (manual setup)

​Authentication

​What you can change

​Clara in nOps vs. Clara MCP

​Troubleshooting

​Stuck on login, or “not authorized”

​Clara doesn’t show up

​Clara responds but the numbers look empty or wrong

​Network

​More help

Clara Slack Agent

Ask Clara

Home page

Integrations

Plugin releases on GitHub

Overview

Connect in 1 minute

Ask your AI agent to set it up

Claude (claude.ai and Claude Desktop)

ChatGPT

Cursor

Start a chat

Good to know

Capabilities

Asking questions about your spend

Exploring spend and building reports

Splitting costs across teams

Budgets, unit economics, and contracts

Commitments and savings

AI and LLM spend

Answers on a schedule

Things to try

Plugins with skills (optional)

Where to get the plugins

Cursor plugin

Steps

Verify

What the Cursor package includes

Global install (all projects)

Claude Desktop plugin

Steps

Verify

What the Claude package includes

Claude Code plugin (CLI)

Other MCP clients (manual setup)

Authentication

What you can change

Clara in nOps vs. Clara MCP

Troubleshooting

Stuck on login, or “not authorized”

Clara doesn’t show up

Clara responds but the numbers look empty or wrong

Network

More help