predator-mcp

An MCP server that exposes the Predator API as tools, so an AI assistant can author load tests, run them, and read the results.

Full documentation: MCP Server in the devguide.

Connect from Claude Code

cd mcp && npm install
claude mcp add predator --env PREDATOR_URL=http://localhost:8088/v1 -- node /path/to/predator/mcp/server.js

PREDATOR_URL is the API root of a running Predator instance (defaults to http://localhost:8088/v1).

Tools

| Tool | What it does | |---|---| | list_tests / get_test | browse test definitions | | create_test / update_test / delete_test | author tests (full artillery v2 scripts — http, kafka, or mixed) | | run_test | launch a run now; returns the report_id | | stop_run | stop a running test | | list_reports / get_report | run status, latency/rps/error aggregates, and kafka consumer-lag over time | | kafka_topics / kafka_consumer_groups | discover a cluster (optional brokers override) | | get_config | read Predator's runtime configuration |

Example prompts

• "Create a kafka test that produces to the orders topic on kafka:9092 and monitors the billing consumer group's lag, then run it for 2 minutes at 50 rps and summarize the report."
• "List my tests, run the checkout one, and tell me if p95 got worse than the last run."