AWS Cost Explorer MCP server: 46 costly tools

AWS Cost Explorer's replacement server has 46 tools for billing questions, with a context cost that exceeds many larger connectors.

Checked 1 September 2026 · from AWS Labs' repository, PyPI and the server's own tools/list response

One of 91 guides to monitoring and developer tools. Each covers what an AI coding agent can get out of the tool through MCP, the protocol that lets the agent query the tool directly.

What is the AWS Cost Explorer MCP server?

It is AWS's server for the Model Context Protocol. What the agent gains is a way to query your AWS spend: what it cost, what it will cost, where it is anomalous and what could be cheaper.

The 46 tools group into four kinds. Reporting reads historical and forecast cost. Optimization reads recommendations from Compute Optimizer, Storage Lens, Cost Optimization Hub and Savings Plans. Governance covers budgets and anomaly monitors. Pricing reads AWS's public price list, which needs no account data at all.

It is read-shaped: nothing here buys a reservation, edits a budget or changes a setting. Two Savings Plans tools can start analysis jobs in Cost Explorer.

It covers cloud cost broadly, and reaches nothing in code review, release, production observability or agent observability.

For an agent, that means AWS answers what your infrastructure cost and what would cost less, and nothing about why. An agent connected here can find the service whose spend jumped and the reservation that would have covered it. It cannot tell you the deploy that caused the jump, so this connector sits beside a release and telemetry one.

What can an AI agent do with AWS Cost Explorer?

What the tool definitions cost in context

A client that loads every tool definition at the start of a session spends this much context before the first question. Claude Code defers definitions by default through tool search, and loads a tool's definition when the agent needs it. The method below gives the version and date measured.

A context window is the amount of text a model can hold at once. A client that loads every tool definition upfront sends each tool's name, description and input schema to the model when the session starts. That list is charged whether or not a single tool is called.

Rootly, Vantage, PagerDuty and Grafana with every category enabled are the only connectors here that cost more. On 0.0.33, 38 tools cost about 26,390 tokens. That is more than Grafana's default 65 tools cost on v1.1.0, and more than GitHub's 44 cost on v1.9.0.

The cause is description length, at 1,867 characters per tool. Those descriptions carry real usage guidance, including which cost metric to prefer and how to shape a query, which is useful and is charged whenever the client loads the tool list, used or not.

These figures were measured on 0.0.33 on 24 August 2026. Version 0.0.39 registers eight more tools, including three for Savings Plans, so its definitions cost more.

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 awslabs.billing-cost-management-mcp-server 0.0.33. It was run locally with placeholder credentials.

Registration is static, so no AWS 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 24%. Reasoning tokens are output, spent per request. The tool list is billed as input whenever the client loads it.

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 close to two fifths.

AWS Cost Explorer MCP server limits: where answers come back incomplete

The limits below are properties of AWS billing data rather than of the connector.

How to configure AWS Cost Explorer MCP for agents

1. 1

   Give it billing read access only

   The tools read billing data, apart from two that start Savings Plans analyses. An IAM role scoped to the Billing and Cost Management APIs keeps the agent's reach to your bill.

2. 2

   Attach it for a session

   At about 26,390 tokens on 0.0.33, and more on 0.0.39, this is among the heaviest connectors in the guide, so it is a poor default and a good investigation tool.

3. 3

   Set the region

   Cost Explorer is a global service reached through a region, and AWS_REGION decides which endpoint the SDK uses.

4. 4

   Prefer a profile over inline keys

   AWS_PROFILE keeps credentials in the SDK's own store rather than in an MCP client's config file.

Paste this into your agent’s instructions

Billing data settles over time and optimization services return nothing when they were never enabled. Both look like ordinary answers, so the block makes the agent qualify them.

When you answer using AWS billing MCP tools, state:
- The exact period you queried, and which cost metric.
  Unblended is the default unless you were asked otherwise.
- That recent days are estimates. Billing data settles, and a
  figure for this week is not the invoice.
- Whether the optimization services were returning data. An
  empty recommendation list may mean they were never enabled.
Never state that data does not exist. State that none was
returned under the constraints 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?

Often you do not. The AWS CLI covers Cost Explorer, Budgets and Pricing with the same credentials, and a coding agent already knows the CLI.

The connector's advantage is the guidance inside the tool descriptions, which is the same thing that made it cost 26,390 tokens on 0.0.33. An agent that reads which cost metric and which dimensions a tool wants composes a better query than one guessing at parameters.

At this size it is a connector to attach for a cost investigation rather than leave connected. Thirteen percent of the window, as measured on 0.0.33, is a lot to spend on a question you are not asking.

Connect it for a billing investigation. Use the CLI when you know the query and want the context back.

AWS Cost Explorer MCP server setup

The package runs with uvx and uses ordinary AWS credentials: a profile the SDK can find, or a key pair, plus a region.

The IAM identity needs access to the Billing and Cost Management APIs. Read access covers every tool except the two that start Savings Plans analyses.

claude mcp add-json "aws-billing" '{"command":"uvx",
  "args":["awslabs.billing-cost-management-mcp-server@latest"],
  "env":{"AWS_PROFILE":"your-profile","AWS_REGION":"us-east-1"}}'

{"mcpServers":{"aws-billing":{"command":"uvx",
  "args":["awslabs.billing-cost-management-mcp-server@latest"],
  "env":{"AWS_PROFILE":"your-profile","AWS_REGION":"us-east-1"}}}}

[mcp_servers.aws-billing]
command = "uvx"
args = ["awslabs.billing-cost-management-mcp-server@latest"]
env = { AWS_PROFILE = "your-profile", AWS_REGION = "us-east-1" }

Every other client

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

{"mcpServers":{"aws-billing":{"command":"uvx",
  "args":["awslabs.billing-cost-management-mcp-server@latest"],
  "env":{"AWS_PROFILE":"your-profile","AWS_REGION":"us-east-1"}}}}

{"mcpServers":{"aws-billing":{"command":"uvx",
  "args":["awslabs.billing-cost-management-mcp-server@latest"],
  "env":{"AWS_PROFILE":"your-profile","AWS_REGION":"us-east-1"}}}}

{"servers":{"aws-billing":{"command":"uvx",
  "args":["awslabs.billing-cost-management-mcp-server@latest"],
  "env":{"AWS_PROFILE":"your-profile","AWS_REGION":"us-east-1"}}}}

{"mcpServers":{"aws-billing":{"command":"uvx",
  "args":["awslabs.billing-cost-management-mcp-server@latest"],
  "env":{"AWS_PROFILE":"your-profile","AWS_REGION":"us-east-1"}}}}

AWS_REGION=us-east-1 \
AWS_ACCESS_KEY_ID=<key> AWS_SECRET_ACCESS_KEY=<secret> \
uvx awslabs.billing-cost-management-mcp-server

Read-only access, permissions and security

• The tools cannot buy or edit

  The connector cannot create a budget, buy a reservation or change a setting. On 0.0.39, sp-recommendation and sp-purchase-analyzer can start analysis jobs in Cost Explorer.

• Billing data is commercially sensitive

  Spend by service, by account and by tag describes your architecture and your scale to anyone reading the conversation.

• Scope the IAM identity to billing

  The tools need Billing and Cost Management APIs and nothing else, so a general-purpose role gives the agent far more reach than it uses.

• Prefer a profile to inline keys

  Credentials then live in the AWS SDK's store rather than in an MCP client's configuration file.

Troubleshooting

Access denied on every tool

The IAM identity is missing Billing and Cost Management permissions. These are separate from general AWS access and often not granted by default.

No optimization recommendations

Compute Optimizer and Cost Optimization Hub return data only after being enabled and collecting history. An empty list is not evidence of an efficient account.

The figure does not match the invoice

Recent billing data is provisional. Usage arrives on a delay and credits apply later, so a figure for the last few days is an estimate.

Pricing looks higher than what you pay

aws-pricing reads the public price list. Enterprise discounts, private pricing and credits are not in it. Use Cost Explorer for actuals.

AWS Cost Explorer MCP server: Reference

What engineers report

This server ships inside the awslabs monorepo, so the useful signals are packaging and history rather than opinions.

Package facts from PyPI, retrieved 6 October 2026. Description lengths and the 0.0.33 count come from the server's own tools/list response on 24 August 2026. The 0.0.39 count comes from the package's tool registrations. The predecessor's removal is recorded in the awslabs monorepo.

Should you connect an agent to AWS Cost Explorer?

Yes. Best for short, deliberate cost investigations, with credentials scoped to billing read access.

Strongest fit

Teams investigating an AWS bill who want an agent to move between spend, anomalies and optimization recommendations in one conversation.

Main advantage

Breadth across billing, and descriptions that teach an agent how to query cost data correctly.

Main weakness

The fifth-heaviest connector here at about 26,390 tokens on 0.0.33, which is a lot to leave attached.

Operational risk

Low. The tools read billing data or start analyses, so the credential is the main exposure.

Frequently asked questions