PocketBase Cloud

PocketBase Cloud MCP Server

The official Model Context Protocol (MCP) server for PocketBase Cloud: a hosted bridge that lets your AI assistant work in your PocketBase Cloud account — projects, PocketBase instances, frontends and backends, deploys and logs, custom domains, and the collections and records inside each instance.

Connect page · Tools · Data and security · PocketBase Cloud

---

PocketBase Cloud hosts PocketBase backends, frontends and server apps. Connect your account once, then ask your assistant in plain words; it picks and runs the right actions.

With the PocketBase Cloud MCP Server, your assistant can:

• See what is running: projects, PocketBase instances, frontends and backends with their status, recent deploys, and the logs of instances and backends.
• Find out why something broke: read a failed deploy's build log and explain the fix.
• Start an app: create projects and PocketBase instances in the region you choose.
• Ship: add and verify custom domains, deploy hooks and static sites, and get the exact pbc command to deploy your source code.
• Work with data: design collections and API rules, and read, create, update and delete records.
• Harden an instance: review API rules for security gaps, find slow requests, and back up before risky changes.

[!IMPORTANT]

You choose the access when you connect: read and write or read only, and all projects or a single one.

Options in the server URL can narrow a connection, never widen it. Deleting needs your explicit

approval of that exact change.

One-click setup

Pick your AI client below. Each button uses your client's native install link, so you don't need to edit any JSON config by hand.

Add to Cursor

Build on PocketBase from your editor.

Add to VS Code

Use it from GitHub Copilot.

Add to Claude

Add it as a custom connector.

Add to ChatGPT

Add it as a custom MCP server.

The server URL, for any other client:

https://mcp.pocketbasecloud.com/mcp

Paste it exactly: the address ends in /mcp.

Let your agent do the setup

Most AI coding agents can add the server and start sign-in themselves. Paste this into your agent:

Add the PocketBase Cloud MCP server https://mcp.pocketbasecloud.com/mcp to this agent

(remote, Streamable HTTP, OAuth), following https://mcp.pocketbasecloud.com/. Then start the authentication flow so I can sign in to PocketBase Cloud.

Claude and ChatGPT

Claude (claude.ai and Claude Desktop), with the plugin. On a paid plan, the plugin brings the skills too:

1. Customize → Plugins, then Add → Add marketplace, and enter pocketbasecloud/mcp.
2. Install PocketBase Cloud.
3. Open the plugin's Connectors tab and connect pocketbasecloud: sign in to PocketBase Cloud, then choose the access and the project. Installing alone does not connect it.

A plugin added there is saved to your Claude account, so Claude Code has it too when you sign in with the same account.

Claude (claude.ai and Claude Desktop), server only

1. Customize → Connectors, then + Add → Add custom connector.
2. Name it (for example *PocketBase Cloud*), paste the server URL, and continue. If asked how Claude identifies itself, choose Register automatically. Click Add.
3. Click Connect, sign in to PocketBase Cloud, then choose the access and the project.

On a Team or Enterprise plan, an owner adds it once under Organization settings → Connectors → Add → Custom → Web; then each member clicks Connect.

ChatGPT

1. Plugins → + → Add custom MCP server.
2. Name it, enter the server URL under Connection with OAuth authentication, accept the warning about custom servers, then Create as a plugin.
3. Install the plugin, sign in to PocketBase Cloud when ChatGPT asks, then type @ in a chat to use it.

Other clients

Scripts and CI: instead of signing in, send a PocketBase Cloud CLI access token (portal → Account → CLI access token) as Authorization: Bearer . See How it works.

Contents

• One-click setup
• Plugin packaging and compatibility
• Tools
• Options
• Before you start
• Data and security
• How it works
• Example prompts
• Skills
• Troubleshooting
• Security
• Support and feedback
• Disclaimer

---

Plugin packaging and compatibility

This repository publishes the same MCP server and skills in several package formats, so each client can use its native discovery and installation flow:

Both mcp.json and .mcp.json are intentional: Agent Plugins requires the root mcp.json and its portable transport vocabulary (streamable-http), while Claude and Cursor read .mcp.json with theirs (http). Gemini CLI names the Streamable HTTP endpoint httpUrl. Keep the endpoint the same in all of them; npm run validate checks it.

Tools

The server groups its tools. Groups marked *opt-in* are added with the features option. The live list, with every tool, is on the connect page.

Not available on purpose: deleting projects or resources (use the portal or pbc), and building source code on the server (it runs on your machine; the server gives you the pbc command).

Options

Add these to the server URL's query string. They can narrow what you approved when signing in, never widen it.

| features | `account,resources