AI Assistant (MCP)

Connect your AI assistant to Costory and explore cloud costs, compare spending, set up alerts, and share reports, all through natural conversation.

Getting Started

Claude Desktop
Claude Code
VS Code (Copilot)
Cursor

Claude Desktop supports MCP connectors natively. No local server or Node.js required.

Open Claude Desktop settings
Go to Connectors
Click Add custom connector
Enter Costory as the name and https://app-api.costory.io/mcp as the URL
Click Add, then Connect
A browser window will open. Log in with your Costory account
Start asking about your costs

Adding Costory as a custom connector in Claude Desktop

Alternative: JSON config

You can also add the server via the developer config file (Developer > Edit Config). This method requires Node.js:

{
  "mcpServers": {
    "costory": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://app-api.costory.io/mcp"]
    }
  }
}

Claude Code connects natively over HTTP with OAuth. No npx or Node.js required.Run this in your terminal:

claude mcp add --transport http costory https://app-api.costory.io/mcp

Then inside Claude Code, run /mcp and select Authenticate. A browser window will open for you to log in with your Costory account.

Alternative: JSON config

You can also add the server directly to your project’s .mcp.json:

{
  "mcpServers": {
    "costory": {
      "type": "http",
      "url": "https://app-api.costory.io/mcp"
    }
  }
}

Add to your .vscode/mcp.json:

{
  "servers": {
    "costory": {
      "type": "http",
      "url": "https://app-api.costory.io/mcp"
    }
  }
}

Alternative: using mcp-remote

If the native HTTP connection doesn’t work, you can use the mcp-remote bridge instead (requires Node.js):

{
  "servers": {
    "costory": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://app-api.costory.io/mcp"]
    }
  }
}

Add to your Cursor MCP settings:

{
  "mcpServers": {
    "costory": {
      "type": "http",
      "url": "https://app-api.costory.io/mcp"
    }
  }
}

Alternative: using mcp-remote

If the native HTTP connection doesn’t work, you can use the mcp-remote bridge instead (requires Node.js):

{
  "mcpServers": {
    "costory": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://app-api.costory.io/mcp"]
    }
  }
}

On first connection, a browser window will open for you to log in with your Costory account. The mcp-remote fallback requires Node.js to be installed.

Why Costory MCP?

Direct access to AWS CUR or a GCP MCP server gives you provider-native line items and dimensions. Costory MCP adds your Costory workspace context so you can ask higher-level cost questions and take action in the same flow.

Query spend by team, product, or feature because , , and are already modeled.
Ask one question across AWS, GCP, Azure, Datadog, Anthropic, OpenAI, and Cursor instead of wiring and maintaining separate MCP servers per provider.
Explain why costs moved by using with deploys, Datadog metrics, and provider events in the same investigation.
Continue from saved views, dashboards, team structures, alert history, and prior investigations so your assistant has persistent workspace context.
Connect with one URL and OAuth (no IAM credentials, Docker, or local server setup), and use it in Claude, Cursor, VS Code, or Claude Code.
Take action directly from chat by creating alerts, saving views, logging events, and sending Slack reports.
Use pre-computed allocation rules and shared costs so each query starts from resolved context instead of rebuilding logic from raw CUR data.

What You Can Do

Explore Costs

Ask questions in natural language and the assistant will query your cost data:

“How much is the payments team spending on compute this month?”
“Show me infrastructure costs for the recommendation engine, broken down by service”
“What’s our total AI spend across Anthropic, OpenAI, and Bedrock?”
“How much are we spending on storage?”

Compare Periods

Understand how costs have changed over time:

“What’s driving the cost increase on our data pipeline this week? Any related deploys or traffic changes?”
“Did the March 15th release affect our search service costs?”
“Compare this week’s spend to last week for the growth team”
“How have our compute costs changed compared to last month?”

Find Information

Search across your entire Costory workspace:

“Find the saved view for the data team’s Kubernetes costs”
“What dashboards do we have for AI spend across providers?”
“Show me alert history related to BigQuery pipelines”

Get Recommendations

The assistant can suggest what to look at:

“What dimension should I group by to explain the checkout cost spike?”
“Based on deploy and metric history, what should I investigate next on the data pipeline increase?”

Take Action

Go beyond analysis. Create views, alerts, events, and share reports:

“Save this query as a view called ‘Recommendation Engine Infra Cost’”
“Set up an alert if payments team compute costs go above $10,000 this month”
“Log an event: deployed search ranking v3 on March 15th”
“Send this cross-provider AI cost report to the #finops Slack channel”

Example Conversations

Investigating a cost spike

You: “Why did our costs go up last week?”The assistant will:

Query cost differences between last week and the week before
Show which services and dimensions drove the change
Check correlated events in the same period (deploys, metric changes, provider events)
Highlight likely causes with a timeline of what changed first
Suggest follow-up actions (drill deeper, set up an alert, etc.)

Cross-provider AI cost review

You: “What’s our total AI spend across Anthropic, OpenAI, and Bedrock this month? Break it down by team and model.”The assistant will:

Query AI-related spend across Anthropic, OpenAI, and AWS Bedrock
Combine results into one total with a provider-level breakdown
Break down the same spend by team and model
Offer to save the view or send the report to Slack

Building a cost report

You: “Show me our top 5 services by cost this month, then send it to #cloud-costs on Slack”The assistant will:

Query costs grouped by service for the current month
Save the query as a view
Send a formatted report with chart to the Slack channel

Setting up monitoring

You: “Create an alert if our daily compute costs go above $5,000”The assistant will:

Create a saved view for compute costs
Set up a cost alert with the threshold
Return a link to review and edit the alert in Costory

Available Tools

Tool	What it does
search	Find dimensions, dashboards, saved views, events, and metrics in your workspace
query_costs	Query cost data with grouping, filters, and time ranges
get_cost_diff	Compare costs between two periods to see what changed
get	Fetch full details of a dashboard, saved view, or advanced explorer
suggest_groupby	Get recommendations for the best dimension to analyze
create_saved_view	Save a cost query for reuse
create_alert	Set up cost threshold alerts with notifications
create_event	Log events to annotate cost changes
list_events	View logged events in a date range
list_alerts	View all active cost and budget alerts
send_to_slack	Send a cost report to a Slack channel
list_slack_channels	See available Slack channels
list_organizations	See which organizations you have access to

Tips

Be specific about time

“last 30 days”, “this quarter”, “March 2026”. The assistant understands natural date ranges.

Use your own names

Mention service names, team names, or project names. The assistant searches your workspace to find the right dimensions.

Ask follow-ups

After a query, drill deeper: “now filter that to just eu-west-1” or “break that down by SKU”.

Multi-org support

If you belong to multiple organizations, just mention which one: “show costs for acme-prod”.

Troubleshooting

Browser doesn't open for login

Claude Code: 'does not support dynamic client registration'

Make sure you’re using the native HTTP transport (type: "http") rather than mcp-remote. Run /mcp in Claude Code, then select Authenticate to trigger the OAuth flow.

'Not authenticated' errors

Your session may have expired. Restart the MCP server in your AI client to re-authenticate. In Claude Code, run /mcp and select Authenticate.

'Multiple organizations' error

Specify which org you want in your message: “show costs for [your-org-slug]”.

No results from search

Try shorter, simpler search terms, e.g. “kubernetes” instead of “kubernetes production dashboard”.

Get Started

Setup

Explore

Tag & Allocate

Reporting & Dashboards

Cost Changes

Teams

Planning

Technical Documentation

Glossary

Getting Started

Why Costory MCP?

What You Can Do

Explore Costs

Compare Periods

Find Information

Get Recommendations

Take Action

Example Conversations

Available Tools

Tips

Be specific about time

Use your own names

Ask follow-ups

Multi-org support

Troubleshooting

Get Started

Setup

Explore

Tag & Allocate

Reporting & Dashboards

Cost Changes

Teams

Planning

Technical Documentation

Glossary

​Getting Started

​Why Costory MCP?

​What You Can Do

​Explore Costs

​Compare Periods

​Find Information

​Get Recommendations

​Take Action

​Example Conversations

​Available Tools

​Tips

Be specific about time

Use your own names

Ask follow-ups

Multi-org support

​Troubleshooting

Getting Started

Why Costory MCP?

What You Can Do

Explore Costs

Compare Periods

Find Information

Get Recommendations

Take Action

Example Conversations

Available Tools

Tips

Troubleshooting