Let your agent set you up

Your coding agent can set wsp up for you and open your first thread. It uses the wsp tools and the wsp skill, which is a file that tells the agent how to use wsp. You give the agent both with one command.

Before you start, install wsp (on a Mac or on Linux) so that wsp runs in your terminal.

Give your agent the wsp tools

1. Go to the folder of the project that you want to work on.

2. Run wsp mcp install with the agent's id:

   cd ~/code/acme
   wsp mcp install --agent claude
   

   sh

3. Restart the agent, so that it loads the new tools.

--agent takes the id of any agent in the catalog, which is the list of agents that wsp knows. For example, it takes claude, codex, gemini, opencode or cursor. To set up more than one agent, give --agent again for each one. If one agent fails, the others still get the tools.

The catalog has an MCP configuration for claude, codex, gemini and opencode. For any other agent, such as Cursor, the command writes the skill and the wsp section, but not the MCP server. For Cursor, the skill goes to ~/.cursor/skills/wsp/SKILL.md. To give that agent the tools, add an MCP server in the agent that runs wsp mcp.

For claude, codex, gemini and opencode, the command writes three things:

If you run the command again, it replaces the wsp section and leaves the other lines of the file as they are. To take the section out, run wsp mcp install --agent <id> --remove. That leaves the MCP configuration and the skill in place.

The last line that the command prints tells you what to say to the agent next.

Ask the agent to set you up

Start the agent in the same folder, and ask it to set you up. In Claude Code, type:

/wsp set up wsp for me

In another agent, type:

Use the wsp skill and set up wsp for me.

The agent follows the setup steps in the skill. It stops at the first step whose result is not what the skill expects, and tells you.

1. It runs wsp --version and wsp threads --json. If no host runs, the second command starts one.
2. It records your project with wsp add.
3. It opens your first thread with wsp run. The thread shows in your sidebar, and you can read and answer it there.

The agent can also add a computer for you. It runs wsp add user@host, and you pick what goes on the computer in the app. Add your first computer shows those steps.

Until the agent restarts, it can do the same work with the wsp command line.

What stays with you

Two things are yours, and the agent cannot do them for you:

• Sign-ins. When a tool must sign in, the agent gives you a page and a code. You finish the sign-in in your own browser.
• Keys. If you pay Anthropic by the token, write ANTHROPIC_API_KEY=<key> in ~/.wsp/.env yourself. On a Claude subscription, you sign in instead. Do not paste a key into the conversation. The host reads the key from that file.

A sudo password also stays with you. If a computer's sudo asks for one, the agent tells you where to type it: in a terminal or in the app.

Start work from your agent

After the setup, ask the agent for work in plain words. For example:

Use the wsp skill and open a thread on acme that adds a dark mode to the settings page.

The agent starts the thread with wsp run, and the thread shows in your sidebar. Coordinating threads from an agent tells you more.