Octopus Deploy MCP Server

The Octopus Deploy MCP Server connects AI assistants such as Claude Code or Cursor to Octopus Deploy, a platform for continuous delivery. Instead of clicking through screens, you ask the assistant which projects exist in a space, what state a deployment is in, or which changes show up in the audit log. The server comes from the vendor itself and, according to the documentation, runs directly inside your Octopus instance, so no local installation and no version management is needed.

What is the Octopus Deploy MCP Server?

Octopus Deploy controls how software is delivered to environments such as test, staging, and production. The MCP server makes this data accessible to assistants. MCP is an open standard through which an assistant calls the tools of an external service. According to the vendor the remote server is reachable at the address of your instance with the path /mcp and requires the instance to be reachable over HTTPS. There is also an older local server as an open-source project on GitHub under the Mozilla Public License 2.0. Its repository is now marked as deprecated; according to the documentation only the remote server receives new features.

Which tools are available?

According to the documentation the remote server offers dedicated tools for common tasks, such as finding spaces and projects. In addition there are four general tools for selected operations of the REST API: one searches for matching operations, one describes them, one runs queries, and one runs commands that change things. According to the vendor the scope initially covers selected core areas, including spaces, projects, tenants, audit events, deployment health, and checking tenant variables. The scope is meant to grow but is smaller than with the local server, which additionally covered releases, runbooks, task logs, and Kubernetes features.

What is it particularly useful for?

The value lies in troubleshooting and overview. If you want to know why a deployment is stuck, which tenants have not set a certain variable, or who changed a configuration last, you get answers in conversation without building filters in the web interface. For platform teams that manage many projects and customer tenants, this saves time on recurring checks. Compared with a script you build yourself on the REST API, the server takes over describing the operations for the assistant and follows the same permissions as the API.

Setup

You need an Octopus instance and an API key. The documentation recommends an agent API key for a dedicated service account, so the assistant has its own identity with narrow permissions and its activity is recognizable in the audit log. The key is passed in the HTTP header X-Octopus-ApiKey. According to the vendor OAuth is planned but not yet available. In Claude Code a command with claude mcp add --transport http is enough; in other clients an entry of type http with address and header.

Requirements, security, and limits

The same security model applies as for the REST API: the assistant may only do what the account behind the key may do. Because one tool can run commands that change things, you should give the service account only the permissions the purpose requires and use a read-only account for pure analysis. In the settings, administrators can switch the remote server off under "MCP Controls". According to the documentation, blocking the local server relies on an identifier of the client that an agent can forge, so it is not a hard boundary. Retrieved data enters the AI client context and therefore reaches the model provider.

Costs

The server is part of Octopus Deploy and requires an instance. According to the pricing page, Octopus Cloud and Octopus Server each offer the Free, Professional, and Enterprise plans. The Free plan is permanently free but limited in projects, tenants, machines, and users; there is also a time-limited trial with all Enterprise features. For current terms see the official Octopus Deploy pricing page.

Who is the Octopus Deploy MCP Server for?

It is worthwhile for teams that already deliver with Octopus Deploy and want to speed up overview and troubleshooting with an assistant. Anyone who wants to create new releases or start runbooks currently finds these functions in the remote server only through the general REST API tools, not as dedicated tools. Anyone who does not use Octopus gains nothing: the server is an access path to the platform, not a standalone tool. The local server's repository is marked as deprecated and therefore only suits existing setups.

An Octopus Deploy instance (Cloud or Server) reachable over HTTPS, an API key, and an MCP client that supports remote servers, such as Claude Code or Cursor.

With the Octopus Deploy MCP Server, an AI assistant queries projects, deployments, and tenants of your Octopus instance. This guide leads from the API key to the first query.

Step 1: Check the prerequisites

You need an Octopus instance that is reachable over HTTPS and an MCP client that supports remote servers, such as Claude Code or Cursor. According to the documentation the server runs inside the instance itself; no local installation is needed.

Step 2: Create a service account and API key

Create a dedicated service account for the assistant and generate an agent API key for it. That way the assistant has its own identity and its activity is recognizable in the audit log. Grant only the permissions your questions require.

Step 3: Register the server

The endpoint is https://<your-instance>/mcp. In Claude Code you register it with claude mcp add --transport http octopus-deploy <endpoint> --header "X-Octopus-ApiKey: ${OCTOPUS_API_KEY}". Keep the key in an environment variable, never in the repository. According to the vendor OAuth is not yet available.

claude mcp add --transport http octopus-deploy <endpoint> --header "X-Octopus-ApiKey: ${OCTOPUS_API_KEY}"

Step 4: First queries

Start with questions such as "Which spaces and projects exist?" or "What is the state of the latest deployments?". Then have audit events checked or find tenants that are missing a variable.

Step 5: Check permissions and switches

Commands that change things run through a dedicated tool. Use an account with read permissions for pure analysis. Administrators can switch the remote server off in the settings under "MCP Controls".

Note on the local server

The repository of the older local server is marked as deprecated and, according to the documentation, receives no new features. For new setups the remote server is the right path.

API key in the HTTP header X-Octopus-ApiKey, recommended as an agent API key of a dedicated service account. OAuth is planned according to the vendor.

The permissions of the account behind the key apply, as with the REST API; read queries and commands that change things are separate tools.

The server runs inside your Octopus instance. Retrieved data enters the AI client context and therefore reaches the model provider.

Unwanted commands that change things with overly broad permissions, passing operational data to the model provider, and an API key stored in the repository.