CloudZero MCP server: 15 tools, 3,000 calls a day

A cost connector that publishes its own quota, caps how much each tool may return, and treats OAuth and API keys differently.

Last verified 1 September 2026 · from CloudZero's own MCP Server Reference documentation

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 CloudZero MCP server?

It is CloudZero's server for the Model Context Protocol. CloudZero allocates cloud spend to teams, products and features, so a connected agent reaches cost as the business sees it rather than as the provider bills it.

The important tool is get_cost_data, with get_available_dimensions and get_dimension_values to find what can be grouped or filtered. set_org_context and get_org_context scope a session to an organization.

CostFormation is CloudZero's allocation configuration, and two tools read it: get_costformation_vers and download_costformation. That is how an agent can see why spend was attributed the way it was.

CloudZero gives an agent cost allocations, CostFormation data, optimization queries and resource telemetry, subject to its published daily quota. It is a cloud-finance connection, not a source of code, pipeline or model-trace evidence.

For an agent, that means CloudZero answers who spent what and why it was attributed that way. The CostFormation tools are the part a provider console cannot give you. It says nothing about how a service behaves, so a telemetry connector belongs beside it.

What can an AI agent do with CloudZero?

set_org_context comes first in any multi-organization account, and get_org_context_versions shows what that context has been.

What connecting costs before the first question

A context window is the amount of text a model can hold at once. Fifteen tools cost 1,460 to 1,642 tokens on Claude, one of the smaller surfaces measured here for a cost platform.

Every tool sits behind a named permission set. The context you pay for and the data you get back are therefore two separate questions.

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 from the 15 tool entries in the vendor's own documentation. That documentation is the only public inventory.

The server could not be run for a live tools/list. No input schemas are available, so these figures cover each tool's name and description only. They are a floor, not the figure.

The Claude columns are measured through the API's count_tokens endpoint with an empty schema per tool. So they cover the same content as the GPT columns. The range there is two tokenizer generations. Haiku 4.5 and Opus 4.6 are at the low end. Opus 5 and Sonnet 5 are at the high end. The gap between them is about a tenth.

CloudZero MCP server limits: where answers come back incomplete

The limits below are documented ones, which is unusual enough to be the reason to read them.

How to configure CloudZero MCP for agents

1. 1

   Prefer OAuth if you will use it heavily

   API-key calls are metered at 3,000 per UTC day per organization; OAuth users are not metered.

2. 2

   Expect caps on result size

   Query-style tools enforce a maximum limit, and a request above the cap is rejected before the tool runs.

3. 3

   Set the organization context

   set_org_context writes session state, which every later cost question depends on.

4. 4

   Check the permission sets

   CloudZero documents data access controls and permission sets that decide what the connection can see.

Paste this into your agent’s instructions

Cost figures depend on an allocation configuration the reader never sees, and results are capped. Both belong in the answer.

When you answer using CloudZero MCP tools, state:
- The organization context and the period the numbers cover.
- Whether the result was capped by a limit, and whether a
  quota warning was appended to the response.
- Whether a figure is allocated cost, an optimization
  recommendation, or a realized saving. These are different tools.
Never state that data does not exist. State that no cost data
matched the context and filters 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?

Yes, for the attribution question. For a scheduled export, the CloudZero API costs no context and brings no quota surprises.

The connector earns its place on the allocation question. Asking why spend landed on a team, and reading the CostFormation that decided it, is awkward anywhere else.

For provider-native billing rather than allocation, AWS Cost Explorer answers what the invoice says, and Vantage covers spend across providers.

Connect it to ask why spend was attributed a certain way. Use the API for scheduled reporting.

CloudZero MCP server setup

The server is hosted, so nothing is installed. CloudZero documents both OAuth and API-key authentication, and they are metered differently.

A Claude Code plugin marketplace entry ships the server pre-configured, which is the shortest route for that client.

claude mcp add --transport http cloudzero \
  https://czca-server.discovery.cloudzero.com/mcp

{"mcpServers":{"cloudzero":{
  "url":"https://czca-server.discovery.cloudzero.com/mcp"}}}

[mcp_servers.cloudzero]
url = "https://czca-server.discovery.cloudzero.com/mcp"

Every other client

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

{"mcpServers":{"cloudzero":{"url":"https://czca-server.discovery.cloudzero.com/mcp"}}}

{"mcpServers":{"cloudzero":{"url":"https://czca-server.discovery.cloudzero.com/mcp"}}}

{"servers":{"cloudzero":{"url":"https://czca-server.discovery.cloudzero.com/mcp"}}}

{"mcpServers":{"cloudzero":{"url":"https://czca-server.discovery.cloudzero.com/mcp"}}}

{"mcpServers":{"cloudzero":{
  "url":"https://czca-server.discovery.cloudzero.com/mcp",
  "headers":{"Authorization":"Bearer <your-api-key>"}}}}

Read-only access, permissions and security

• Almost everything reads

  Of the 15 documented tools, only the context tools change anything, and what they change is session scope.

• Permission sets gate the data

  CloudZero documents data access controls and permission sets separately from the tool list.

• Quota exhaustion is visible

  As an organization approaches the daily limit the server appends a warning to responses, so it degrades loudly rather than silently.

• Response size is bounded

  A documented response size limit and per-tool limit caps stop a single call returning an unbounded result.

Troubleshooting

Calls stop working partway through a day

API-key calls are capped at 3,000 per UTC calendar day per organization, resetting at midnight UTC. OAuth users are not metered.

A request is rejected before the tool runs

Query-style tools cap the limit parameter. get_available_dimensions allows up to 1,000, and requests above a tool's cap are refused.

Cost figures cover the wrong account

Call set_org_context first. In a multi-organization account, the context decides what every later question means.

The numbers disagree with the provider console

CloudZero allocates spend using CostFormation. Read download_costformation to see the configuration that produced the attribution.

CloudZero MCP server: Reference

What engineers report

CloudZero publishes operational limits most vendors leave unstated, so the figures below are its own.

Tool inventory, quotas and limit caps read from CloudZero's own MCP Server Reference on 24 August 2026. An unauthenticated tools/list against the endpoint returned 401 on the same date, which is why this page is sourced from documentation rather than from the wire.

Should you connect an agent to CloudZero?

Yes. Best for OAuth from an interactive client, with set_org_context called at the start of every session.

Strongest fit

CloudZero customers who want an agent to explain attribution rather than restate a provider invoice.

Main advantage

Published quotas, permission sets and per-tool caps, which makes capacity planning possible before you connect.

Main weakness

API-key use is metered at 3,000 calls a day, which an exploratory agent can consume quickly.

Operational risk

Low. Almost every tool reads, and the only writes are session context.

Frequently asked questions