mcp-k6

The k6 MCP Server (mcp-k6) is an experimental MCP implementation that integrates with k6 to provide script validation, test execution, documentation navigation, and guided script generation. It exposes tools to validate k6 scripts with minimal run configuration, execute performance tests with configurable load profiles, browse official k6 documentation in a structured way, and generate production-ready k6 scripts from natural language requirements. You can connect editors or MCP clients to the server via the supported transport modes (stdio by default or streamable HTTP in HTTP mode), enabling local or remote usage depending on your workflow.

Key capabilities include: script validation (validate_script) to catch syntax or runtime issues early; test execution (run_script) to run performance tests locally with specified VUs, durations, and stages; documentation browsing (list_sections and get_documentation) to explore k6 docs and retrieve markdown for sections; and script generation (generate_script) to produce production-grade k6 tests from plain-English requirements using embedded best practices and type definitions. This makes it easier to iterate on tests, verify correctness, and maintain consistent scripting standards across your team.

Prerequisites:

• Docker installed on your host (recommended for quick setup)
• Internet access to pull the container image

Installation steps (Docker method):

1. Pull the official image:

docker pull grafana/mcp-k6:latest

2. Run the MCP server in Docker (HTTP transport is optional; default is stdio):

# Basic run (stdio transport by default)
docker run -i grafana/mcp-k6:latest

# Run with HTTP transport enabled (example on port 8080)
docker run -p 8080:8080 grafana/mcp-k6:latest -transport=http -addr=:8080

3. Verify the server is responding (for HTTP mode, you can query the MCP endpoint at http://localhost:8080/mcp):

curl http://localhost:8080/mcp

If you prefer native or other install methods described in the project, you can follow the README's options for Homebrew, native Go installation, or package installs, but the Docker approach is the quickest to get started.

Notes and tips:

• The project is experimental; expect edge cases and provide feedback to the maintainers.
• The HTTP transport mode can be used for remote deployments or team shared instances; when using HTTP, be mindful of security since the server may not include built-in authentication.
• Editor integrations typically point to the MCP endpoint over stdio or HTTP depending on setup; ensure your editor configuration matches the chosen transport.
• If you run in HTTP mode, you might want to place the container behind a secure proxy or VPN for access control.
• For updates, pull the latest image or rebuild your local install if you’re using a native or package-based setup.