Introduction

Webdock MCP lets compatible AI assistants work directly with your Webdock account.

Instead of copying server details into a chat and then manually carrying out every instruction, you can give an AI assistant access to Webdock tools. Depending on the permissions you approve, the assistant can inspect your infrastructure, provision servers, manage services, run supported operations, and help with multi-step infrastructure tasks.

MCP Endpoint: https://mcp.webdock.io

What is MCP?

MCP stands for Model Context Protocol. It is an open standard that lets AI applications connect to external tools and services in a consistent way.

A simple way to think about MCP is as a standard interface between your AI assistant and Webdock.

Without MCP:

You ask an AI how to do something.
It gives you commands or instructions.
You carry them out yourself.

With Webdock MCP:

You ask the AI what you want to accomplish.
The AI discovers the relevant Webdock tools.
It can inspect the current state and, when allowed, perform the required Webdock actions for you.

Webdock MCP uses OAuth. You sign in to Webdock in your browser and choose which permissions the AI client may use. You do not need to paste your Webdock password or a general-purpose API token into the AI client.

What can I use it for?

The exact tools available can change as the Webdock API evolves, but typical workflows include:

• inspecting servers, profiles, locations, images and account information;
• creating and configuring servers;
• choosing an appropriate server profile and software image;
• using Webdock’s LAMP or LEMP images when your application needs a ready web stack;
• creating snapshots before risky changes;
• resizing or reconfiguring servers;
• managing supported WordPress functionality;
• managing MySQL databases;
• running supported server-side operations and scripts;
• diagnosing problems by inspecting infrastructure state before making changes;
• handling multi-step workflows where one operation must finish before the next begins.

The tools exposed to your AI assistant are limited by the Webdock permissions you approve during authorization.

Connect your AI client

Webdock is a remote MCP server using HTTP and OAuth.

In an MCP client that supports remote servers, the basic setup is usually:

1. Add a new custom or remote MCP server.
2. Choose HTTP or Streamable HTTP if the client asks for a transport.
3. Enter: https://mcp.webdock.io
4. Choose OAuth authentication, or simply connect if the client discovers OAuth automatically.
5. Sign in to Webdock in the browser window that opens.
6. Review and approve the requested permissions.
7. Return to the AI client and verify that Webdock tools are available.

Permissions: start with what you need

Webdock MCP permissions control what the connected AI client can do on your behalf.

For an assistant you only want to use for inventory and troubleshooting, read access may be enough. For an assistant that should provision, reconfigure or operate servers, it will need additional permissions.

As a general rule, grant the permissions needed for the work you expect the assistant to perform.

Some clients are better than others at requesting extra permissions later. Devin and Visual Studio Code are currently notable cases where scope escalation can be unreliable. Webdock therefore pre-selects the available permissions for these clients during authorization. You can deselect permissions, but if you expect the client to manage your infrastructure broadly, leaving the required permissions selected avoids tasks failing later because the client did not request a missing permission.

Give the AI a good goal, not just a command

MCP works best when you describe the outcome you want and the important constraints.

Instead of:

Create a server.

Try:

Create an inexpensive server in Denmark for a small PHP application. Use a Webdock LEMP image so Nginx, PHP and the database stack are already available. Show me the profile and monthly price before provisioning anything billable.

That gives the agent enough context to choose better tools and make fewer assumptions.

Use Webdock’s LAMP and LEMP images when they fit

If you already know your application needs a traditional web stack, tell the agent.

Webdock provides ready LAMP and LEMP server images. These can be a better starting point than asking the AI to provision a bare operating system and then manually assemble the same stack.

For example:

I am deploying a PHP application that needs Nginx, PHP and MySQL. Prefer Webdock’s LEMP image instead of a bare OS image.

or:

Build this on Webdock’s LAMP stack unless there is a strong reason not to.

The same principle applies to managed capabilities. If you specifically want Webdock’s managed WordPress tooling, say so. The agent can inspect profile capabilities and choose a compatible profile before provisioning.

Ask for a snapshot before large or risky changes

AI assistants can make mistakes, and perfectly valid changes can also have unexpected application-level consequences.

Before major changes to an existing server, consider explicitly asking the agent to create a snapshot first.

For example:

Before changing the PHP version, web-server configuration or database packages, create a snapshot and wait for it to complete. Then make the changes and verify the site is healthy.

This is particularly useful before:

• major package upgrades;
• web-server or PHP configuration changes;
• database migrations;
• large application deployments;
• destructive cleanup;
• changes where rollback would otherwise be difficult.

A snapshot should not replace your normal backup strategy, but it can provide a useful recovery point before infrastructure changes.

Tell the agent to wait for asynchronous operations

Some Webdock operations are asynchronous. A request being accepted does not necessarily mean the operation has finished.

For multi-step work, a good instruction is:

Wait for each Webdock operation to finish successfully and verify the resulting state before continuing to the next step.

This is especially useful during provisioning, snapshots, deletion, reconfiguration and other longer-running operations.

Use a capable model

Infrastructure management is not a good place to optimize too aggressively for the smallest or cheapest AI model.

The model needs to:

• select the right tools;
• understand dependencies between operations;
• interpret errors and preconditions;
• keep track of server IDs, profiles and returned state;
• avoid unsafe shortcuts;
• verify that asynchronous operations actually finished.

In our testing, lightweight models performed noticeably worse on multi-step infrastructure work. We saw a lightweight Haiku variant make poor choices and tool mistakes that stronger models handled correctly.

For meaningful server administration, use a capable general-purpose or reasoning model. A smaller model may be fine for simple read-only questions, but we do not recommend it for autonomous infrastructure changes.

Good prompting habits

A few small instructions can make agent-driven infrastructure work much safer and more predictable.

Ask it to inspect before changing

Inspect the current server configuration and explain what you plan to change before making modifications.

Ask it to prefer native Webdock capabilities

Prefer Webdock API/MCP functionality over manual shell workarounds when an appropriate Webdock-managed operation exists.

State your priorities

Choose the cheapest profile that supports Webdock-managed WordPress.

is different from:

Choose the absolute cheapest VPS. I am happy to install and maintain WordPress manually.

Ask it to show billable choices first

Tell me the profile, displayed monthly price and location before creating the server.

Name the exact server for dangerous work

Only perform this on server my-production-server. Do not modify any other server.

Ask for verification

After the change, re-read the server state and verify that the requested result actually took effect.

Tell it when not to improvise

If a Webdock-managed operation is unavailable because of a plan or profile limitation, stop and explain the options instead of working around it manually.

Example prompts

Create an inexpensive managed WordPress server

Create a new inexpensive WordPress server in Denmark. I want to use Webdock’s managed WordPress tooling, so first inspect available profiles and choose the cheapest compatible one. Use a suitable Webdock LEMP image. Show me the monthly price before provisioning, then install WordPress and set the site title to “Hello World”.

Prepare a safe application upgrade

Inspect my server example-server and tell me what stack it is running. Before making any major package or configuration changes, create a snapshot and wait for it to finish. Then upgrade the application dependencies, verify the site is responding normally, and summarize exactly what changed.

Find a suitable profile

I need a server in Denmark with at least 4 GB RAM. Show me the cheapest profiles that satisfy that requirement, including the human-readable monthly price and included managed services. Do not provision anything yet.

Use a ready web stack

Create a small PHP server in Denmark. Prefer Webdock’s LEMP image rather than a bare Ubuntu image because I want Nginx, PHP and database components ready for the application. Show the proposed server configuration before creating it.

Investigate without changing anything

Inspect my Webdock servers and identify anything that looks unusual. This is read-only: do not restart, resize, delete or modify anything.

Make a risky change with a recovery point

I need to make a major configuration change on example-server. Create a snapshot first and verify it completed. Then explain the change you intend to make and ask me before executing the risky step.

Remember: the AI is an operator, not an oracle

MCP gives an AI assistant useful tools, but it does not make the model infallible.

For production systems:

• use a capable model;
• grant permissions deliberately;
• be explicit about constraints;
• review billable and destructive operations;
• create recovery points before major changes;
• ask the agent to verify the final state.

With those habits, Webdock MCP can turn a capable AI assistant into a very effective infrastructure operator while keeping Webdock account permissions under your control.

Please reach out to Webdock Support if you need assistance.