Skip to Content

Agent Skills Setup

Before you begin

The skills need two things to do real work:

  1. The CompetLab MCP server configured with your API key. This is how every skill reaches your data. If you haven’t set it up, follow the Connect guide first — it’s the hard prerequisite. Without it, a skill tells you it can’t run. A read key is enough for the whole suite.
  2. A CompetLab project with competitors being monitored — for six of the seven skills. They read your monitoring data, so you need an account and at least one project set up. New here? Create a free account  and set up your first project first. A trial works. The exception is competlab-site-audit, which runs on any public domain with no project and no competitors configured — a free-trial key is enough.

Authentication lives entirely in the MCP configuration. There’s no separate skill-level API key to set.

Install

Pick whichever path fits your setup — all three install the same seven skills and work identically. Every skill carries what it needs inside its own folder, so nothing depends on which path you chose.

In a recent version of Claude Code, add the marketplace and install the plugin:

/plugin marketplace add competlab/competlab-ci-skills /plugin install competlab-ci-skills@competlab-ci-skills

The install string is plugin@marketplace — both happen to be named competlab-ci-skills. This is the cleanest path for Claude Code: it keeps the skills updatable through the marketplace.

The suite is released on GitHub; this page describes v3.3.2 .

Confirm the MCP connection

The skills are only as useful as the MCP connection behind them. Before running one, confirm your agent can reach CompetLab — ask it to list your projects, or check that the mcp__competlab__* tools are available. If the connection isn’t there, the skills will say so rather than invent data. (Details in the Connect guide.)

Run your first skill

Skills trigger on plain-language requests — each one advertises the phrasings it responds to, so you don’t memorize commands. A few examples:

Are we in the core? AI visibility report for my Acme project. Why are they recommended and not us? Catch me up on what changed with competitors. Write a battlecard for Rival Inc. Deep dive on competitor rival.com. Audit example.com for agent adoption. Am I tracking the right competitors?

Your agent recognizes the intent, loads the matching skill, pulls the data through the MCP server, and writes the deliverable. For a single-competitor skill like the battlecard or the dossier, name the competitor; for the site audit, name the domain; for the rest, name the project if you have more than one.

Try it with no project

competlab-site-audit is the one skill that works before a project exists. It runs on any public domain with no project and no competitors configured — crawler access, sitemap coverage, agent adoption, tech stack, trust signals, and whether a claimed MCP server answers the protocol or only a browser. It still reads through the authenticated MCP server, so a CompetLab API key is required; a free-trial key is enough.

Audit example.com for agent adoption.

Other agents

The skills are written for Claude Code, and the tool names inside them use Claude Code’s mcp__competlab__* convention. Cursor, Codex, Gemini CLI, and any agent that reads the SKILL.md format can run them too — you may need to adjust the tool names to match how your client namespaces MCP tools.

Next steps

  • Skill reference → — all seven skills, what each reads, and the phrasings that trigger them.
  • MCP Connect → — set up or double-check the MCP server the skills depend on.
Last updated on