Quickstart

ContextPie is an MCP server your agent reads at runtime. Add the endpoint once, give your agent a one-line rule to search it — from then on your agent discovers and loads your skills on its own, and every call is logged so you can see what actually got used.

Your skills live in a workspace, and each workspace has its own MCP endpoint. A default workspace is created for you on sign-up, so you can start here and read Workspaces when you're ready to organize teams or products.

1. Create a skill

Sign in and create your first skill in the editor. A skill is a SKILL.md file plus any supporting docs, scripts, or templates. The two fields that matter most for discovery are the description and the trigger phrases — they're what get embedded and searched.

Spend your effort on the description and trigger phrases. They decide whether a skill is ever found. See Authoring a skill.

2. Add the endpoint

Copy your workspace's MCP endpoint from the Connect screen — it ends in /w/{your-workspace}, so the connection only sees that workspace's skills. The fastest path in Claude Code is the CLI:

claude mcp add --transport http contextpie https://mcp.contextpie.com/w/your-workspace

Or add it to your client's MCP config directly (ContextPie uses the streamable HTTP transport):

{
  "mcpServers": {
    "contextpie": {
      "type": "http",
      "url": "https://mcp.contextpie.com/w/your-workspace"
    }
  }
}

Restart the agent. ContextPie's three tools — find_skills, get_skill, and get_skill_file — now appear. See Connecting an agent for per-client snippets.

3. Teach your agent to use it

Adding the server only makes the tools available — it won't make the agent reach for them. Add a short rule to your agent's instructions so it searches ContextPie on its own instead of waiting to be asked. For Claude Code that's CLAUDE.md; other clients have their own file (Cursor's .cursor/rules, Copilot's .github/copilot-instructions.md, or AGENTS.md).

## Using ContextPie skills
Skills for this project are served by the ContextPie MCP server ("contextpie").
Before starting a non-trivial task, call `find_skills` with a short description of
what you're about to do. If a result is relevant, load it with `get_skill` and
follow it — fetch any referenced files with `get_skill_file`. Prefer a matching
skill over improvising, and check proactively without being asked.

The Connect screen has a copyable version tailored to your client.

4. Use your agent normally

Ask for something one of your skills covers. Your agent calls find_skills, loads the best match with get_skill, and fetches referenced files on demand with get_skill_file. You don't orchestrate any of this — the agent does it through progressive disclosure.

5. Read the signal

Open the Dashboard. Because the agent's queries flow through ContextPie, every search, load, and file fetch is captured as a byproduct of serving the request — the query is the telemetry. You'll see what's used, what's dead, and where discovery failed. More in Analytics.