Vantage MCP server: 127 tools across cloud cost

What a cost platform exposes to an agent, what the tool list costs before any query, and which tools change your billing configuration.

Last verified 1 September 2026 · from Vantage's repository, npm and the server's own tools/list response

This page is one of 90 in a directory of monitoring and developer tools. Each page checks what an AI coding agent can get out of the tool through its MCP server, the connector that lets the agent query the tool directly.

What is the Vantage MCP server?

It is Vantage's server for the Model Context Protocol. Vantage is a cloud cost platform, so a connected agent reaches spend across providers rather than one cloud's own billing console.

The tools mirror the Vantage API's resources: cost reports, budgets, anomalies, billing rules, folders, dashboards, workspaces, and virtual tag configurations, which re-map spend onto labels of your own. query-costs is the general entry point and takes VQL, Vantage's own query language.

Scoping matters more here than on most connectors. query-costs must be scoped to a workspace, and get-myself is how an agent finds which workspaces the token can see.

Vantage sits in cost management, and reaches nothing in code review, CI or agent observability.

In practice, Vantage answers what things cost and where the money went. It can attribute spend, compare periods and open a budget alert. It cannot tell you why a service got slower or which deploy caused it, so pair it with a telemetry connector.

What can an AI agent do with Vantage?

Call get-myself first. It returns the workspaces the token can reach, and query-costs refuses to run without one.

What connecting costs before the first question

A context window is the amount of text a model can hold at once, and connecting uses some of it before the agent asks anything. Every tool the server registers sends its name, description and input schema to the model when the session starts. The tool list is charged whether or not a single tool is called.

127 tools cost between 36,433 and 65,575 tokens, which is 18.2% to 32.8% of a 200,000-token window.

Both halves contribute. Descriptions average 413 characters and schemas average 815, so this is a wide surface described at ordinary length rather than a few tools with enormous schemas.

Encodings: GPT-5.6, GPT-5.5 o200k_base; GPT-4, GPT-3.5 cl100k_base; Claude Haiku 4.5 to Opus 5 count_tokens. Counted 24 August 2026 from tools/list on the released npm package 2.24.0. It was run locally with a placeholder token.

Registration is static, so no Vantage account was involved. Figures cover name, description and input schema.

The two models in column one share the o200k_base encoding, so they get one count. The spread is formatting, about 40%. Reasoning tokens are output, spent per request. The tool list is input, charged once at connection.

The Claude figures are measured rather than encoded. Anthropic publishes no local tokenizer. So the definitions were sent to the API's count_tokens endpoint, and a no-tools baseline was subtracted. The range covers two tokenizer generations rather than two serializations. Haiku 4.5 and Opus 4.6 sit at the low end. Opus 5 and Sonnet 5 sit at the high end. The gap between them is about a third.

Vantage MCP server limits: where answers come back incomplete

The limits below come from the breadth of the surface and from writing tools arriving on by default.

How to configure Vantage MCP for agents

1. 1

   Scope the token before connecting

   All 127 tools arrive together and 57 of them write, so what the token may do is the whole boundary.

2. 2

   Decide who owns billing rules

   create-billing-rule can exclude, credit or charge cost data, which changes what every report shows afterwards.

3. 3

   Set the workspace expectation

   query-costs requires a workspace scope, so an agent that skips get-myself will fail its first query.

4. 4

   Budget the context

   At 36,433 tokens on the low measurement, this connector takes 18.2% of a 200,000-token window before any question.

Paste this into your agent’s instructions

Cost answers look authoritative and are easy to misattribute, and several tools quietly change how attribution works. The block makes the agent name both the scope and any change.

When you answer using Vantage MCP tools, state:
- The workspace you queried and the period the numbers cover.
- Whether a number is actual spend, a forecast, or a
  recommendation. The three are different tools.
- Whether you called a create-, update- or delete- tool, and
  on what. Billing rules and virtual tags change every report.
Never state that data does not exist. State that none was
returned under the scope above.
Write your answer in ASD-STE100 Simplified Technical
English. Use short sentences with one idea in each.

Do you need the MCP server at all?

Sometimes. Vantage publishes a REST API, and for a single known report a direct call skips the 36,433-token charge entirely.

The connector earns its place on exploration. query-costs takes VQL, so an agent can shape a question it did not know the endpoint for, then follow it into anomalies and recommendations.

For one cloud only, that provider's own cost connector is narrower and cheaper. Vantage is worth the context when the spend is spread across providers.

Call the API for a known report. Connect the server when the agent has to find where the money went.

Vantage MCP server setup

The package runs with npx and needs one Vantage API token in VANTAGE_TOKEN. The server exits at startup if it is missing.

Vantage also documents a hosted OAuth mode, which suits a team better than sharing one token in a config file.

claude mcp add-json "vantage" '{"command":"npx",
  "args":["-y","vantage-mcp-server"],
  "env":{"VANTAGE_TOKEN":"<your-token>"}}'

{"mcpServers":{"vantage":{"command":"npx",
  "args":["-y","vantage-mcp-server"],
  "env":{"VANTAGE_TOKEN":"<your-token>"}}}}

[mcp_servers.vantage]
command = "npx"
args = ["-y", "vantage-mcp-server"]
env = { VANTAGE_TOKEN = "<your-token>" }

Every other client

Each block below is the configuration for one client, with the file path and the key that client expects.

{"mcpServers":{"vantage":{"command":"npx",
  "args":["-y","vantage-mcp-server"],
  "env":{"VANTAGE_TOKEN":"<your-token>"}}}}

{"mcpServers":{"vantage":{"command":"npx",
  "args":["-y","vantage-mcp-server"],
  "env":{"VANTAGE_TOKEN":"<your-token>"}}}}

{"servers":{"vantage":{"command":"npx",
  "args":["-y","vantage-mcp-server"],
  "env":{"VANTAGE_TOKEN":"<your-token>"}}}}

{"mcpServers":{"vantage":{"command":"npx",
  "args":["-y","vantage-mcp-server"],
  "env":{"VANTAGE_TOKEN":"<your-token>"}}}}

# Issue a read-scoped Vantage token for the agent,
# then use it exactly as above.
VANTAGE_TOKEN=<read-scoped-token> npx -y vantage-mcp-server

Read-only access, permissions and security

• 57 of the 127 tools write

  The names say so: create-, update- and delete- prefixes across reports, budgets, billing rules, dashboards and virtual tags.

• One delete announces itself as irreversible

  delete-cost-report states in its own description that the action cannot be undone.

• Billing rules rewrite what reports show

  create-billing-rule can exclude, credit, charge or otherwise modify cost data, so a change there moves every downstream number.

• There is no read-only mode

  The boundary has to be the API token's scope, issued for the agent rather than shared with a person.

Troubleshooting

The server exits immediately

VANTAGE_TOKEN is not set. The server prints "VANTAGE_TOKEN environment variable is required" and stops before registering any tool.

A cost query is rejected

query-costs must be scoped to a workspace. Have the agent call get-myself first to find which workspaces the token can reach.

Numbers moved without anyone changing the cloud

Check whether a billing rule or virtual tag configuration changed. Both are writable over this connector and both re-attribute existing spend.

The agent uses a large share of context on connection

The definitions measure 36,433 to 65,575 tokens. Filter the tool list in the client if your window is tight.

Vantage MCP server: Reference

What engineers report

The repository is actively maintained: 84 stars, MIT licensed, last pushed 19 August 2026 and released as 2.24.0 on npm the same day. The repository, the package and the measurement carry the evidence rather than public discussion.

“When trying to use this server with cursor, it fails to load any tools.”

Repository facts from the GitHub API and npm, retrieved 24 August 2026. Tool counts, descriptions and schema sizes come from the server's own tools/list response on the released package.

Should you connect an agent to Vantage?

Yes. Best for a read-scoped token issued for the agent, on a client that can filter the tool list when the window is tight.

Strongest fit

Teams whose spend spans several cloud providers and who want an agent to explore it rather than open a console.

Main advantage

A complete cost surface in one connector, with VQL for questions no fixed endpoint answers.

Main weakness

The definitions take 18.2% to 32.8% of a 200,000-token window before any question is asked.

Operational risk

57 writing tools arrive enabled, including deletes and billing rules that re-attribute existing spend.

Frequently asked questions