Distributed Load Testing on AWS MCP
An MCP server from the AWS load testing solution: an agent reads scenarios and runs and starts JMeter, K6 and Locust tests
- 423 ★ on the repository
- Last commit
Broad access
Why this level
- Starts and stops load tests that send traffic from Fargate containers to the target service
- The server writes and deletes scenarios and scripts in the S3 bucket of your AWS account
- The load is billed to your account: a mistake in task or Region counts raises the bill
Install
- Works in
Manual install
make deployDeployment from source; fill in the environment variables first, details are in the Local Development section.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Starts and stops load tests that send traffic from Fargate containers to the target service
- The server writes and deletes scenarios and scripts in the S3 bucket of your AWS account
- The load is billed to your account: a mistake in task or Region counts raises the bill
This is third-party code. Review the repository files before installing.
What it does
Distributed Load Testing on AWS (DLT) is a ready-made AWS Solutions package for load testing. ECS containers on Fargate generate traffic from several Regions, and scenarios are written in JMeter, K6 or Locust or defined with one HTTP endpoint. A run can start later or repeat on a schedule. An agent needs one optional part of this repository: the MCP server for analyzing tests. The server is written in TypeScript and runs as a Lambda function behind Amazon Bedrock AgentCore Gateway. An MCP client calls the gateway address, the gateway validates a Cognito token and forwards the call to Lambda, which calls the solution's REST API and the S3 bucket that holds scenarios. The agent reads scenarios, runs, the baseline run and artifacts. `source/mcp-server/src/tools` also holds write tools: creating, updating, starting and stopping tests, schedules and script upload. A scenario runs in one of two traffic-shape modes. In Standard mode the solution sets the Fargate task count, virtual users and load profile through Taurus. In Native mode your script controls the load: the solution runs the k6, Locust or JMeter file with no load flags, and the only dials are the task count per Region and a required safety duration of up to 24 hours. The `get_workflow_guides` tool gives the agent step-by-step workflows, including `create_native_and_run`.
Who it is for. Performance and DevOps engineers on AWS teams who run DLT or are ready to deploy it and want an agent to help read load test results.
Good fit when
- An agent has to compare the latest load test run with the baseline and name what got worse
- Results of dozens of scenarios sit in the DLT console, and you would rather ask questions in plain language
- A k6 or Locust script from CI has to be uploaded and run as is in Native mode
- A schedule of nightly runs is easier to create or change from a chat with the agent
Not a fit when
- DLT is not deployed in your AWS account: the server does not work without it
- You did not pick the MCP Server option at deployment: the component is installed only with it
- The load would hit someone else's service without the owner's permission: such a test harms third parties
Example request
Compare the latest run of the checkout-api scenario with the baseline and show where response time grewMCP
- Transport
- http
- Authentication
- OAuth
Limitations
- A DLT stack in your AWS account is required. The MCP server is an optional component of it and is deployed only when you pick the MCP Server option.
- Access goes through Amazon Bedrock AgentCore Gateway with a Cognito token check, so the MCP client must be able to obtain the user's token.
- The Fargate containers that generate load run in, and are billed to, your AWS account: more tasks and Regions mean a bigger bill.
- The documentation targets architects and administrators with AWS experience. GitHub lists the repository license as NOASSERTION, so read the LICENSE file for the terms.
How to disable. Disable the MCP server in the agent's client and delete the gateway connection settings. To remove the component from your AWS account, delete the DLT stack in CloudFormation.
FAQ
How do I connect DLT to Claude Code or Cursor?
Deploy DLT with the MCP Server option, then put the AgentCore Gateway address and the user's Cognito token into the MCP client configuration. There is no ready-made address: the gateway is created in your AWS account.
Which load frameworks does DLT support?
JMeter, K6 and Locust, plus a simple setup for one HTTP endpoint. The simple endpoint works only in Standard mode, while Native mode needs an uploaded script.
How does Native mode differ from Standard?
In Standard, DLT sets users, ramp-up and hold through Taurus. In Native, the script drives the load and DLT keeps only the task count and a time limit. A script with 200 users on five tasks produces 1,000 users.
What can an agent do through this MCP server?
Read scenarios, runs, the baseline run and artifacts. According to the source it can also create, update, start and stop tests, manage schedules and upload scripts. Your MCP client lists the full set of tools.
Where does the server run and who can use it?
In your AWS account, as a Lambda function behind AgentCore Gateway. The gateway lets through only calls with a valid Cognito token from the solution's user pool.
Discussion
No questions or reviews yet.
Related
MCP serversA Rust HTTP load-testing tool with an MCP server: an agent runs a benchmark and gets latency percentiles and status breakdowns.
MCP server: an agent sends GET, POST, PUT, DELETE and PATCH to your REST API and gets back status, headers and body
kogiQA MCP gives an agent a browser that finds page elements without selectors, for debugging web apps and writing e2e tests
MCP server, agent and skills for end-to-end UI tests where the agent drives the computer through vision on cloud machines
MCP server for a device test bench: an agent leases a board, connects and runs jmp commands on real or virtual hardware.
MCP Interviewer checks an MCP server before an agent connects: provider limits, tool tests and a Markdown report.
README in short
The page describes the solution and how it works: a CloudFront console or a headless option, Cognito sign-in, Fargate task launches and metrics collection. A separate section explains the two traffic-shape modes. It then covers deployment templates for CloudFormation and CDK, the source layout and local development.