run-mcpjam-evals

name	run-mcpjam-evals
description	Drive MCPJam's hosted eval tools end to end — check what a run will cost and disclose, launch it, poll it to a verdict, and triage a failure down to the step that failed. Use when connected to MCPJam's MCP server and asked to run, re-run, investigate, or compare eval results, rather than to author eval files locally.

run-mcpjam-evals

You have MCPJam's hosted eval tools. This is the order to call them in, and the two places calling them wrong costs real money.

This skill is about running evals that already exist. To write eval files, use create-mcp-eval (SDK tests) or mcpjam-eval-import (convert an existing corpus). To author suites through these tools instead, see references/authoring.md.

Reference map

You are about to…	Read
Work out why a run did not pass	references/triage.md
Create a suite or add cases through the tools	references/authoring.md

The loop

list_eval_suites → get_eval_run_disclosure → run_eval_suite → get_eval_run (poll)
                                                                    ↓
                                          list_eval_run_iterations → get_eval_run_steps
                                                                    ↓
                                                            compare_eval_run

1. Orient. list_eval_suites returns suites with latest-run summaries and pass-rate trends. With no project it uses the most recently updated accessible one — fine for a quick look, but pass project explicitly the moment you are about to change or spend anything, or you may act on a project the user did not mean.