MCPLab

App

MCPLab Assistant Skill

Install and use the mcplab-assistant skill from skills.sh in Codex/Claude-style agent workflows.

What This Skill Is For

The mcplab-assistant skill is an operator guide for running MCPLab workflows through an AI coding assistant. It focuses on practical operations: authoring eval config YAML, running commands, triaging failures, and interpreting run artifacts.

Use it when you want consistent, deterministic help with MCPLab usage rather than generic advice.

  • Config authoring and edits for MCPLab eval files.
  • Command help for `mcplab run`, `mcplab app`, and `mcplab report`.
  • Failure triage for auth/config/selection/numeric-flag errors.
  • Output analysis for `results.json`, `summary.md`, `trace.jsonl`, and `report.html`.

Install via Skills CLI

Install the skill with the Skills CLI from the source repository. This pattern is useful before or independent of a leaderboard slug.

install mcplab-assistant
npx skills add https://github.com/inspectr-hq/mcplab --skill mcplab-assistant

Use in Agent Chats

After installation, ask your coding agent to use the skill explicitly when working with MCPLab. Keep prompts task-focused and include file paths or command output when debugging.

config authoring
Use the mcplab-assistant skill to draft a minimal eval config with one scenario and OAuth client-credentials auth.
run + compare
Use mcplab-assistant to run this config and compare claude-haiku vs gpt-4o-mini with --agents.
failure triage
My run fails with fetch failed. Use mcplab-assistant to triage. Here is the command, stderr, and relevant config block.
result analysis
Use mcplab-assistant to analyze this run directory and summarize why failing scenarios failed.

Verify Installation and Usage

A successful setup should satisfy all checks below.

  • The install command completes without errors.
  • Your agent can confirm the `mcplab-assistant` skill is available.
  • Prompts that request the skill return MCPLab-specific, command-ready guidance.
  • Guidance follows a deterministic structure: Intent -> Actions -> Verification -> If It Fails.

Troubleshooting

  • Install failed: verify the GitHub URL is reachable and the skill name is exactly `mcplab-assistant`.
  • Skill not detected by your agent: restart the agent session and re-check installed skills.
  • Guidance is generic: explicitly request use of `mcplab-assistant` and provide config snippets/command output.
  • Command mismatch: prioritize current MCPLab CLI contracts from local `packages/cli/src/cli.ts`.