W14: Introducing MCP

Eric Hubbell

Overview
This week we introduced @playbooks/mcp, a new MCP server for Playbooks.
It gives frontier models and coding agents a direct way to work with Playbooks from tools like Claude Code, Cursor, Codex, and VS Code.
This matters because the Playbooks interface is only part of the workflow. As agents become more involved in how developers explore, download, publish, and deploy software, Playbooks needs a first-class entrypoint for that layer too.
What's New
- Added
@playbooks/mcpas a new MCP server for Playbooks - Added support for clients like Claude Code, Cursor, Codex, and VS Code
- Added CLI helpers for installing MCP client configuration
- Reused the same Playbooks session and config model as
@playbooks/cli - Exposed common Playbooks workflows to agents through MCP tools
Built on the CLI
@playbooks/mcp does not reimplement Playbooks logic from scratch.
Instead, it wraps the Playbooks CLI so agents can use the same account, config, and project workflows that already exist at the terminal layer.
That keeps the model simpler and gives us one shared path for authentication, account access, and local project actions.
What Agents Can Do
Through MCP, agents can help with common Playbooks workflows such as:
- checking authentication and session state
- exploring marketplace and account resources
- downloading, adding, and cloning plays into local projects
- syncing, publishing, and deploying plays
The goal is not just chat-based access. It is to make Playbooks practical inside real agent-driven workflows.
Getting Started
The fastest way to connect a client is through the CLI helpers:
Terminalplaybooks mcp claude playbooks mcp cursor playbooks mcp codex playbooks mcp vscode
From there, developers can authenticate with the CLI and start using Playbooks through a fresh MCP-enabled chat.
That's all for now
MCP is an important step toward making Playbooks usable anywhere developers and agents work together. We’re excited to keep building both the CLI and MCP layers so the platform feels just as natural in an agent workflow as it does in the browser.


