Appwrite MCP Server

The Appwrite MCP Server connects AI assistants such as Claude Code or Cursor to your Appwrite project. Appwrite is an open-source backend platform that offers databases, user accounts, file storage, serverless functions, and website hosting under one roof. The MCP server makes these features callable for an assistant: instead of clicking through the console, you ask the assistant to create a collection, list users, or check the status of a function. It can also search the official Appwrite documentation, so answers rest on the current docs rather than outdated training knowledge. The project lives in the GitHub organization appwrite, is released under the MIT license, currently has around 70 stars, is not archived, and was last updated on October 23, 2026.

What can the Appwrite MCP Server do?

According to the documentation the server provides four tools. appwrite_get_context summarizes your account, organization, and projects. appwrite_search_tools searches Appwrite's full tool catalog at runtime, and appwrite_call_tool calls a specific tool by name. With appwrite_search_docs the assistant searches the documentation semantically. This design keeps the assistant's context small: it does not load hundreds of API functions up front but looks up what it needs. According to the README this gives access to the Appwrite API with databases, users, functions, teams, storage, and more. The documentation also describes an events extension: agents can subscribe to changes in the project, such as newly created users, failed function executions, or completed deployments, instead of polling.

Hosted with OAuth or local?

The default route is the hosted server at https://mcp.appwrite.io/. It uses HTTP as transport and OAuth for sign-in. On first connection your client opens a consent page from Appwrite where you approve the requested permissions; you do not need to copy API keys. An Appwrite Cloud project is required. Anyone running Appwrite themselves starts the server locally over stdio instead and authenticates with a project API key; the setup is described in a separate guide in the repository. For clients without remote server support the README names a workaround using the helper mcp-remote, which passes the remote server through over stdio.

What is it particularly useful for?

The greatest value lies in the development phase. While building an app you have the assistant create tables and permissions, check test data, or find out why a function failed, without leaving the editor. The documentation search helps with detail questions about SDKs and configuration. Compared with a general database server, the Appwrite MCP Server differs in that it knows the whole backend platform, including user management, functions, and storage, and in that sign-in runs through OAuth with explicitly approved permissions.

Security: what to keep in mind

An assistant with write permissions can change or delete data, users, and functions. On the consent page, check carefully which permissions you grant, and work in a test or development project first rather than in production. For the local variant the same applies to the API key: grant only the permissions needed. Retrieved content enters the AI client's context and therefore reaches the model provider; this includes user data the assistant reads. For the events feature the documentation names limits: each subscription or open stream uses one project webhook, which is two per project on the free plan, and events during a reconnect are not replayed. We did not check the full list of tools individually; the official documentation is authoritative.

Costs

The MCP server itself is free under the MIT license. For hosted operation you need an Appwrite Cloud project. According to the pricing page there is a free plan in which the project is frozen when it reaches its resource limits, and a paid Pro plan in which projects keep running and extra usage is billed; a budget cap can be set. Current terms are on Appwrite's official pricing page.

Who is the Appwrite MCP Server for?

It is worthwhile for developers who already work with Appwrite or are evaluating a backend platform for a new app and want their assistant to work directly on the project. Without an Appwrite project it adds little. Start with the hosted server and a test project, use read-oriented tasks such as context and documentation search first, and enable write permissions only once you know the flow.

Hosted: Appwrite Cloud project and an MCP client that supports HTTP servers with OAuth. Self-hosted: your own Appwrite instance, a project API key, and a local start over stdio.

With the Appwrite MCP Server an AI assistant operates your Appwrite project and searches the documentation. This guide leads from sign-in to the first task.

Step 1: Prerequisites

You need an Appwrite Cloud project and an MCP client that supports HTTP servers with OAuth, such as Claude Code or Cursor. Anyone running Appwrite themselves uses the local variant and a project API key.

Step 2: Connect the hosted server

In Claude Code the command is claude mcp add --transport http appwrite https://mcp.appwrite.io/. Then open /mcp in a session, select appwrite, and follow the prompt in the browser. On Appwrite's consent page you approve the permissions; you do not copy API keys.

claude mcp add --transport http appwrite https://mcp.appwrite.io/

Clients without remote support

If your client does not support remote servers, the README names a workaround with the helper mcp-remote, which passes the server through over stdio and requires Node.js.

Self-hosted Appwrite

For your own instance you start the server locally over stdio and authenticate with a project API key. The steps are in the self-hosted Appwrite guide in the repository. Give the key only the permissions you need.

Limit the risks

Begin with a test project and check carefully on the consent page what you approve. An assistant with write permissions can change or delete data and users.

Step 3: First tasks

Start with reads: "Summarize my projects" or "Search the documentation for how to set permissions on a table". Only then have collections created or functions checked.

OAuth through an Appwrite consent page in the browser, without API keys; the self-hosted local variant uses a project API key.

Access to the Appwrite account, organization, and projects within the scope of the approved permissions, including databases, users, functions, and storage as well as documentation search; write access is possible depending on the permissions granted.

Retrieved project data, user information, and documentation excerpts enter the AI client context and therefore reach the model provider. Processing by Appwrite follows the provider's privacy policy.

Unintended changes or deletion of data, users, and functions with overly broad permissions, exposure of user data to the model provider, an overly broad API key for self-hosted use, and prompt injection via stored content.