Skip to main content
Spree hosts a Model Context Protocol server on top of this documentation. Once connected, your agent can search the docs and read full pages on demand — answering questions from the latest published documentation instead of stale training data.
The server exposes two tools: full-text search across the documentation, and reading doc pages as Markdown.
This server is about Spree’s documentation, not your store. To let an agent work on your own catalog, orders and settings, see the Admin MCP server — the same protocol, hosted by your application and authenticated with one of your API keys.

Setup

In a create-spree-app project, one command connects Claude Code, Cursor and VS Code at once, and installs the agent skills too:
It adds the server to .mcp.json, .cursor/mcp.json and .vscode/mcp.json, keeping any servers those files already list, and points the project’s AGENTS.md at it. Running it again changes nothing. Commit the files so your team gets the same setup. To connect a single tool by hand, or one outside a create-spree-app project:
Scope it to the project (--scope project) to share the connection with your team via .mcp.json, or keep the default local scope for just yourself.

When to use MCP vs the other doc channels

New create-spree-app projects come connected (skip it with --no-docs-mcp) — Claude Code asks you to approve the server the first time you open the project. The project’s AGENTS.md tells agents to search here and read pages from the local @spree/docs package, which matches the Spree version the project runs.