Use With AI Agents

An agent reaches for whatever tools it has. Until it resolves the contract, it cannot tell which of them this session actually authorizes. PersonaKit prepares that contract; your coding agent does the work. This page connects the two so an agent can discover PersonaKit on its own and resolve the active contract before it starts, instead of waiting for you to paste setup into chat.

It assumes you already have a PersonaKit root. If you do not, run personakit init <path> first, or follow the Start Here walkthrough.

1. Register The MCP Server

The read-only MCP server is the fastest path. Register it once and the agent reaches PersonaKit live, with nothing pasted into the prompt. For Claude Code, one command registers it at user scope, so it is available in every project:

claude mcp add personakit --scope user -- "$(which personakit)" mcp

Confirm it connected:

claude mcp list

You should see personakit listed as connected. Other hosts use the same stdio server (personakit mcp) with their own config format; the per-host table for Cursor, GitHub Copilot, Codex, VS Code, and OpenCode lives in the MCP guide.

Open the per-host MCP setup table

2. Add A Grounding Skill

The MCP server gives the agent access. A small host skill gives it the instinct to reach for PersonaKit at the right moment: resolve the active contract before choosing tools, skills, or files. The skill is only the front door; PersonaKit remains the authority for what is allowed.

personakit init scaffolds a host-neutral personakit-grounding/SKILL.md. Per-host variants for Claude Code, Cursor, GitHub Copilot, and OpenCode are in the grounding tutorial. Copy the variant for your host into the directory it discovers skills from. For Claude Code that is .claude/skills/.

Open the grounding skill examples

3. Point A Cold Agent At PersonaKit

An agent that does not know PersonaKit exists will not look for it. If your project keeps an AGENTS.md, a short breadcrumb tells a cold agent where to orient. This snippet is opt-in; PersonaKit does not scaffold it into your repo.

## PersonaKit

This project uses PersonaKit for operating contracts. Before starting work,
run `personakit guidance` (or read `personakit://catalog/start` when the
PersonaKit MCP server is connected) to resolve the active session contract:
role, rules, allowed capabilities, forbidden actions, and stop points.

What The Agent Does Once Connected

With the server connected and the contract resolved, the agent follows a small, inspectable loop:

1. Orient: read personakit://catalog/start, or run personakit guidance with no MCP server.
2. Resolve: recommend or select a session, then resolve its contract.
3. Respect the map: work inside the cleared space; at a stop point or an unauthorized capability, re-ground or ask instead of improvising.

personakit guidance

Agents consuming PersonaKit from any host can go deeper in the agent guide.