MCP Server

An optional MCP server that lets an AI agent search this documentation on demand, from inside the editor, while it writes code. It runs locally — nothing to host — and by default indexes the latest published docs, so the agent reads the same reference you do.

It is optional: /llms.txt and /llms-full.txt cover the same knowledge statically. Add the MCP server when you want the agent to pull the exact current page for an endpoint mid-task instead of ingesting the whole index.

Why use it

• Live lookup while coding. The agent fetches the precise request/response shape for an endpoint the moment it needs it — no copy-pasting docs into the chat.
• Always current. It indexes the published docs, not a stale snapshot from training data.
• Smaller context. The agent pulls only the page it is working on, leaving room for your code.

Tools it exposes

Tool	What it does
search_api_docs	Ranked full-text search across the API documentation
list_endpoints	Enumerate endpoints, filterable by transport (rest/graphql) and menu
get_doc	Fetch one page's full content by path
refresh_api_docs	Re-pull the latest docs in-session, no restart

Install

git clone https://github.com/bagisto/mcp
cd mcp
npm install

Requires Node 18+. Then register it with your agent — the panel on the right has the exact command for Claude Code, Codex, and Antigravity. Any other MCP client (Cursor, Windsurf, VS Code) uses the same stdio node <path>/src/index.mjs form.

• Repository: bagisto/mcp

Docs source

By default the server fetches https://api-docs.bagisto.com/llms-full.txt and caches it locally. To pin a local snapshot or point at a mirror, set BAGISTO_DOCS_LLMS to a URL or file path when registering. Call refresh_api_docs to pull newly-published docs mid-session.

Do you need it?

No — it is a convenience, not a requirement. With the Agent Skills installed and /llms.txt reachable, you already have everything needed to build. Add the MCP server for on-demand, in-editor doc search on top of that.