Vantage Launches a Remote MCP Server

Today, Vantage is launching a remote version of its Model Context Protocol (MCP) Server, hosted on Cloudflare. Customers can now use Vantage’s MCP functionality, which enables AI agents like Claude, Amazon Bedrock, and Cursor to interact with customers’ cloud cost and usage data, without needing to install packages or manage infrastructure associated with running an MCP.

Vantage launched its self-hosted MCP earlier this year, just months after Anthropic open-sourced the Model Context Protocol. At this time, only self-hosted MCPs were supported, and only integrated with locally installed agents through desktop applications. In May, Anthropic announced support for remote MCPs, and the industry has been coalescing around provider-hosted, remote MCPs as a more frictionless delivery and distribution method. Unlike the local MCP servers, remote MCP servers are accessible on the internet, opening up integration points to more than just desktop AI agents. Users can simply sign in and grant permissions to MCP clients using familiar authorization flows, opening up new ways of working with LLMs and agents to a wider audience.

Now, the Vantage MCP Server is available as a fully managed, hosted service, eliminating the need for customers to install and host their own Vantage MCP server. Customers can connect AI agents, using both desktop and hosted agents, to their cost and usage data without deploying anything themselves. The Remote MCP offers secure transport, built-in OAuth authentication, and zero-effort version upgrades. By simply pointing AI agents to https://mcp.vantage.sh/sse, customers can immediately begin asking cost questions, automating workflows, or embedding the MCP into team tools, like Slackbots or internal copilots.

The Remote Vantage MCP Server is now available for all Vantage customers. To get started, head to the Integrations page or visit the Vantage MCP Documentation for setup instructions.

Frequently Asked Questions

1. What is being launched today?

Vantage is launching a remote version of its MCP Server, which allows customers to use AI agents to access and interact with their organization’s data within Vantage without managing their own MCP deployment.

2. Who is the customer?

This feature is available to all Vantage users, especially those interested in using GenAI tools in their FinOps workflows.

3. How much does this cost?

There is no additional cost for customers to use the Remote Vantage MCP Server.

4. How is the Remote Vantage MCP Server different from the Self-Hosted Vantage MCP Server?

With the Remote Vantage MCP Server, there’s no need to deploy infrastructure, manage API keys, or handle version upgrades. Vantage manages everything for you. Authentication is handled via OAuth, with each request scoped to the authenticated user’s permissions, and prompts are processed in stateless, isolated environments without being stored or visible to Vantage. This provides a secure, scalable, and zero-maintenance way to connect AI agents to your cloud cost data.

5. What functionality does the Remote Vantage MCP Server have?

The Remote MCP offers all the same functionality as the local MCP. At the time of release, v0.0.4 of the Vantage MCPs have the following read-only tools:

• query-costs- a general purpose way to fetch Cost data using VQL.
• list-costs- display all the Costs in an associated Cost Report.
• list-cost-reports- list all Cost Reports available.
• get-cost-report-forecast- list all Forecasts of spending related to a Cost Report.
• list-cost-integrations- list all Cost Provider integrations (e.g., AWS, Azure, GCP) available to provide Costs data from and their associated accounts.
• list-cost-providers- list of just the Providers that the given Workspace has shared with it, for filtering in VQL queries.
• list-cost-services- lists all the Services and their associated Provider that is shared with the given Workspace.
• list-budgets- list all Budgets available to compare against a Cost Report and track spending.
• list-dashboards- list all Dashboards created in the Vantage account.
• list-tags- list Tags that can be used to filter Cost Reports.
• list-tag-values- list Tag values that can be used to filter Cost Reports.
• list-anomalies- list Anomalies that were detected on Cost Reports.
• list-unit-costs- retrieve the Unit Costs for a given Cost Report.
• get-myself- a utility to list available Workspaces and check the access level of your auth token.
• submit-user-feedback - a simple way to send feedback about the MCP or overall Vantage experience to the Vantage team.

For a continually updated list, please see the Vantage MCP documentation.

6. How do users authenticate with the Remote Vantage MCP Server?

Authentication with the Remote MCP is handled via OAuth. Once you add the MCP to your AI agent, you will be prompted in a browser to authenticate with Vantage in the same manner you would authenticate as if logging in to console.vantage.sh.

7. Can I deploy an instance of the Remote MCP that is not associated with a Vantage user?

No. At this time, authentication is all done against existing Vantage users. You can still use the Self-Hosted MCP for authentication of non-Vantage users through an API key.

8. Does Vantage have access to see what prompts I ask an LLM?

Prompts that invoke the Vantage MCP are isolated into the AI agent’s environments, and are not made accessible to Vantage. Vantage can see what APIs are called on the user’s behalf, but has no information into the prompts users submit.

9. What are the supported MCP Clients I can use this with?

The Vantage MCP Server is compatible with any AI agents that provide MCP Client functionality. At the time of this blog post, examples of supported clients include Anthropic’s Claude (desktop application and claude.ai), Azure AI Foundry, Amazon Bedrock, Google Gemini, Cursor IDE, and Block’s local AI agent Goose. See the documentation from each provider to understand minimum version requirements required to run MCP hosts for each, as well as integrate remote MCPs.

10. Can I use the MCP with ChatGPT?

At this time, the official ChatGPT desktop client supports MCP clients only for deep research. General-use MCPs, like Vantage, are not supported. You can use the agents such as Goose to connect to the ChatGPT-powered LLMs, and also connect to the Vantage MCP Server.

11. If I need support using the Vantage MCP Server, how can I get that?

Through our regular support channels. Please email support@vantage.sh and/or reach out to your Customer Success Manager where appropriate.

12. I would like to provide feedback or talk more on this. How can I engage more with the Vantage team on MCP-related work?

In the Vantage Community Slack, there is a new #mcp channel where you can provide feedback and interact with Vantage employees contributing to the project, as well as interact with other Vantage users on how they are utilizing the Vantage MCP Server. The Vantage Community Slack is a free community for anyone to join. Additionally, you can use the submit-user-feedback tool within the MCP to submit feedback by simply telling the MCP to provide Vantage any piece of feedback

13. Are there any limits to the number of calls I can make with the Vantage MCP?

The Vantage MCP Server itself does not impose explicit call limits. However, usage is subject to Vantage’s underlying API rate limits, which are applied at the account level. These limits include a general rate limit of 1,000 requests per hour across all API endpoints, and a more restrictive limit of 5 requests per 5 seconds, specifically for Cost Report endpoints.

It’s important to note that these are account-wide limits rather than per-user restrictions, meaning all API activity within your Vantage account contributes to these thresholds. Additionally, when using the MCP with AI agents, you should consider the token limits and processing constraints of your chosen AI platform, as these may also impact your overall usage patterns and workflow efficiency.

14. Can I revoke my MCP Access Token?

Yes, you can revoke your MCP Access Token at any time in the “API Access Token” page within settings, and then navigating to the “MCP Server Token” tab. If you have an MCP Server Token, you will be able to delete it here.