Configure Remote MCP

Last verified 23 Oct 2026

DigitalOcean provides remote MCP servers that expose tools and context for managing your DigitalOcean resources. You can connect in two ways:

  • The MCP Router, a single endpoint that fronts all of DigitalOcean’s first-party MCP services. This is the recommended approach.
  • Individual service endpoints, one per service, which you add only for the services you need.

In both cases, you add the server URLs and authentication details to your local mcp.json file, and your MCP client uses this configuration to connect to the servers and manage your resources. The location of mcp.json and how it’s applied depend on your MCP client.

Prerequisites

To use the remote MCP servers, you need the following:

Warning

Do not commit access tokens to your Git repository. Many MCP clients require you to put access tokens directly in their configuration files. This is safe only if the configuration file is not checked into a Git repository and has appropriately restricted read permissions.

If the configuration file is in a Git repository, add it to .gitignore or use any available client-specific techniques (for example, VS Code’s input variable substitution) to avoid exposing your access token.

Connect Through the MCP Router

The DigitalOcean MCP Router is a single remote MCP server that fronts all of DigitalOcean’s first-party MCP services. Instead of adding each service’s endpoint separately, you connect once to https://mcp.digitalocean.com/mcp and your MCP client discovers and runs tools across every supported service through it. This is the recommended way to use DigitalOcean’s remote MCP servers.

The router exposes three tools that work together as a discovery-to-run flow:

  • find_tools takes a natural-language query and returns a ranked shortlist of candidate tools, each with its qualified name (in server/tool form) and a short description. An optional limit argument sets the number of candidates, five by default and up to 25.
  • retrieve_schema returns the full schema for one tool by its qualified name: its input schema and an output schema when the tool provides one.
  • invoke_tool runs a tool by its qualified name, passing arguments that match the tool’s input schema. You can optionally set a projection, a list of dot-notation paths such as droplet.id, to return only selected fields of the tool’s structured result. Projection requires the tool to expose an output schema. Paths can’t index into arrays, so select the whole array (for example, droplet.networks.v4) instead of a single element.

Your MCP client calls find_tools to locate a tool, retrieve_schema to get its schema, then invoke_tool to run it. The router forwards your credentials to the service that owns the tool, so you authenticate only once and reach DigitalOcean’s available MCP services.

To use the router, add it to the mcpServers block of your client’s configuration. We recommend authenticating with OAuth, which prompts you to sign in through your browser and stores no long-lived credentials:

{
  "mcpServers": {
    "digitalocean": {
      "url": "https://mcp.digitalocean.com/mcp"
    }
  }
}

You can also authenticate with a DigitalOcean API token by adding an Authorization header. Replace <your_api_token> with your token, or use your client’s variable substitution method to keep the token out of the configuration file:

{
  "mcpServers": {
    "digitalocean": {
      "url": "https://mcp.digitalocean.com/mcp",
      "headers": {
        "Authorization": "Bearer <your_api_token>"
      }
    }
  }
}

Configure Individual Remote MCP Servers

If you want to connect to specific services instead of using the MCP Router, you can add each service’s endpoint individually in mcp.json.

To enable a remote MCP service, add it to the mcpServers block of your client’s configuration. Each remote service entry requires three pieces of information:

  • A unique service key, such as apps or databases
  • The service’s URL, pointing to the hosted MCP server
  • A headers block, providing your authentication (for example, a bearer token)

The following example configures connections to the App Platform (apps) and Databases (databases) remote MCP servers. Replace <your_api_token> with your DigitalOcean API token, or use the client-specific method of variable substitution (if available) to keep your token out of the configuration file.

{
  "mcpServers": {
    "apps": {
      "url": "https://apps.mcp.digitalocean.com/mcp",
      "headers": {
        "Authorization": "Bearer <your_api_token>"
      }
    },
    "databases": {
      "url": "https://databases.mcp.digitalocean.com/mcp",
      "headers": {
        "Authorization": "Bearer <your_api_token>"
      }
    }
  }
}

Authenticate with OAuth

We recommend authenticating with OAuth instead of a static API token. Your client obtains a short-lived access token through a browser-based sign-in, so you don’t store a long-lived token in your configuration. All remote MCP servers that require authentication support OAuth.

To use OAuth, add the server’s URL to your client’s configuration without an Authorization header:

{
  "mcpServers": {
    "apps": {
      "url": "https://apps.mcp.digitalocean.com/mcp"
    }
  }
}

Available Remote MCP Service Endpoints

Product/Service Remote MCP Server URL Service Key Suggestion
App Platform https://apps.mcp.digitalocean.com/mcp apps
Kubernetes https://doks.mcp.digitalocean.com/mcp doks or kubernetes
Droplets https://droplets.mcp.digitalocean.com/mcp droplets
MicroVMs https://microvms.mcp.digitalocean.com/mcp microvms
Functions https://functions.mcp.digitalocean.com/mcp functions
Databases https://databases.mcp.digitalocean.com/mcp databases
Dedicated Inference https://dedicated-inference.mcp.digitalocean.com/mcp dedicated-inference
Model Catalog https://inference-modelcatalog.mcp.digitalocean.com/mcp model-catalog
Knowledge Bases https://kbaas.do-ai.run/v1/mcp knowledge-bases
Spaces https://spaces.mcp.digitalocean.com/mcp spaces
Networking https://networking.mcp.digitalocean.com/mcp networking
Network File Storage https://nfs.mcp.digitalocean.com/mcp nfs
Accounts https://accounts.mcp.digitalocean.com/mcp accounts
Insights https://insights.mcp.digitalocean.com/mcp insights
Evaluations https://genai-evaluation.mcp.digitalocean.com/mcp genai-evaluation
Batch Inference https://genai-batchinference.mcp.digitalocean.com/mcp batchinference
Signals https://signals.mcp.digitalocean.com/mcp signals
Harness Runtime https://harness-runtime.mcp.digitalocean.com/mcp harness-runtime
Custom Models https://genai-custom-models.mcp.digitalocean.com/mcp genai-custom-models
Inference Router https://genai-inferencerouter.mcp.digitalocean.com/mcp genai-inferencerouter
Vector Database https://vector-databases.mcp.digitalocean.com/mcp vector-databases
Container Registry https://docr.mcp.digitalocean.com/mcp docr
Documentation https://docs.mcp.digitalocean.com/mcp docs
Marketplace https://marketplace.mcp.digitalocean.com/mcp marketplace
Volumes Block Storage https://volumes.mcp.digitalocean.com/mcp volumes

The MCP Router fronts all of the servers listed above except the Knowledge Bases server (https://kbaas.do-ai.run/v1/mcp), which you must connect to directly.

The Documentation MCP server (docs) does not require an API token. You can omit the Authorization header when configuring it.

Refer to the mcp-digitalocean repository for more information on supported DigitalOcean services and detailed lists and descriptions of their available tools.

Add Remote MCP Servers to Your MCP Client

Claude

To use DigitalOcean remote MCP servers with Claude Code and Claude Desktop, navigate to Claude Code’s settings and add a custom connector, then provide the remote MCP server URLs.

Remote MCP servers do not use claude mcp add, environment variables, or local executables.

For more information, see the official Claude Support article Getting Started with Custom Connectors Using Remote MCP.

Cursor

See Cursor’s Model Context Protocol docs to add a DigitalOcean remote MCP server to Cursor.

Visual Studio Code

See Use MCP servers in VS Code to add a DigitalOcean remote MCP server to Visual Studio Code.

Windsurf

See the Windsurf Cascade MCP docs to learn how to add a DigitalOcean remote MCP server to Windsurf.

Troubleshooting

If your DigitalOcean MCP server is not working as expected, check the following common issues.

Authentication Issues

The DigitalOcean MCP server requires an active and properly-scoped access token:

  • Ensure the DIGITALOCEAN_API_TOKEN variable is correctly set in the headers block of the MCP server.
  • Verify the access token has appropriate permissions for the services you’re using.
  • Check that the access token hasn’t expired.

Unsupported Region or Service

An LLM may configure services or features that are not available in the specified region:

  • Check regional availability for the service you’re setting up.
  • Verify the service is available in your selected region.
  • Use doctl compute region list to see available regions.

Debugging Best Practices

When debugging issues with LLMs and the DigitalOcean MCP server, consider the following:

  • Keep Logs: Save the MCP request/response logs for reproducibility.
  • Start Small: Test with small and focused operations first before moving on to complex workflows.
  • Check Permissions: Ensure your API token has the necessary scopes.

We can't find any results for your search.

Try using different keywords or simplifying your search terms.