DigitalOcean MCP Server

The DigitalOcean MCP Server connects AI assistants such as Claude or Cursor to the cloud platform DigitalOcean. Instead of clicking through the web interface or typing commands, you phrase a task, for example "Show me all Droplets in Frankfurt" or "Resize the server for the staging environment", and the assistant makes the matching calls to the DigitalOcean interface. According to the repository the project lives under the digitalocean-labs organization, is released under the MIT license, and was last updated on October 19, 2026. Whether DigitalOcean officially supports it as a product is not stated explicitly in the README we checked; however, the documentation on docs.digitalocean.com describes the MCP servers itself, and the hosted endpoints run under the company's own domain.

What is the DigitalOcean MCP Server?

MCP stands for Model Context Protocol, an open standard that lets assistants use tools outside their chat window. The DigitalOcean server provides such tools for 24 services, including apps, Droplets, databases, Kubernetes, networking, object storage (Spaces), volumes, functions, AI services, and a documentation search. Examples from the README are tools for resizing a Droplet or flushing a CDN cache. Internally the server builds on the Go library godo and the mcp-go framework.

Remote or local

There are two modes of operation. The recommended route is the remote servers hosted by DigitalOcean. According to the documentation a single endpoint is enough, the MCP Router at https://mcp.digitalocean.com/mcp, which provides the tools search_tools and invoke_tool and thus reaches all services without an overloaded tool list. If you prefer to connect individual services, use separate addresses such as https://apps.mcp.digitalocean.com/mcp or https://databases.mcp.digitalocean.com/mcp. Alternatively the server runs locally with npx @digitalocean/mcp --services apps,droplets; this requires Node.js 18 or later and npm 8 or later. The README advises enabling only the services you actually need so the model's context stays small.

Setup and sign-in

OAuth is recommended: you enter only the server address, the client opens the browser on first access, and a short-lived token is issued without permanent credentials ending up in a configuration file. Alternatively you pass a personal API token in the Authorization: Bearer header, or through the DIGITALOCEAN_API_TOKEN environment variable with the local server. In Claude, remote servers are added as a custom connector in the settings, according to the documentation. For Cursor, VS Code, and Windsurf the page points to their own documentation. The token needs permissions for the services in use; the page names no specific scopes and refers to the repository for them. The documentation search service works without a token.

What is it particularly useful for?

Its value lies in routine tasks and in finding your way around an account: listing inventory, redeploying apps, inspecting databases and clusters, checking network and storage resources. Compared with a pure documentation server this one can actually change resources; compared with the command line it saves looking up commands and parameters. For small teams without a dedicated platform department it is a convenient shortcut, though one to use with care.

Security

If you give the assistant a token with write permissions, you also allow it to shrink servers, change databases, or delete resources. Start with a few services, use a token with the narrowest possible rights or OAuth, and review every destructive action before confirming it. Tokens do not belong in Git repositories.

Costs

According to the pricing page and the documentation the MCP server is not billed separately; the pricing page names no fee of its own for it. What can incur costs are the resources the assistant creates or enlarges. DigitalOcean bills many products by time and some by usage, and according to the pricing page new accounts receive a time-limited starting credit. Current terms are on the official pricing page.

Who is the DigitalOcean MCP Server for?

It is worthwhile for developers and small teams that already run on DigitalOcean and want to handle recurring administration tasks in conversation. Without a DigitalOcean account it is useless. A catalog entry also does not replace your own review: try the server first in a test account or with read-only tasks.

A DigitalOcean account and an MCP client that supports remote servers, such as Claude or Cursor. The local server needs Node.js 18 or later, npm 8 or later, and an API token.

With the DigitalOcean MCP Server, an AI assistant manages Droplets, apps, and databases. This guide leads from the prerequisites to the first query.

Step 1: Check the prerequisites

You need a DigitalOcean account and an MCP client that supports remote servers, such as Claude or Cursor. According to the documentation the recommended endpoint is the MCP Router at https://mcp.digitalocean.com/mcp. Only Knowledge Bases must be connected directly.

Step 2: Set up in Claude

Open the connectors in the settings, add a custom connector with the address of the MCP Router, and confirm the sign-in in the browser. A short-lived token is issued and no permanent credentials are stored.

Step 3: Set up in Cursor

In the mcp.json file, add an entry digitalocean with the address as url and sign in via OAuth on first access. Alternatively add a headers block with Authorization: Bearer and your token. Do not put this file in a Git repository then.

Limit services

Instead of the router you can connect individual services, such as https://apps.mcp.digitalocean.com/mcp or https://databases.mcp.digitalocean.com/mcp. With the local server you choose services with npx @digitalocean/mcp --services apps,droplets and set DIGITALOCEAN_API_TOKEN. Node.js 18 or later is required for this.

npx @digitalocean/mcp --services apps,droplets

Step 4: First queries

Start with read tasks, such as "List my Droplets" or "Which apps are running in my account?". Spot-check the answers in the DigitalOcean interface. Only then is it worth using write access, and you confirm destructive actions deliberately each time.

OAuth 2.0 via browser sign-in (recommended) or a personal API token in the Authorization header or in DIGITALOCEAN_API_TOKEN with the local server.

Read and write access to DigitalOcean resources within the rights of the account or token; the scope can be limited by choosing which services are enabled.

The remote servers run hosted by DigitalOcean. Retrieved resource and account data enters the AI client context and therefore reaches the model provider.

Unintended changes or deletions of servers, databases, and storage through write permissions, ongoing costs from newly created or enlarged resources, and an API token stored in the repository.