Argo CD MCP Server: Manage Deployments with AI Agents
Jiacheng Xu
Originally published May, 2025. Updated October, 2026
September 2026 Update: Akuity launched the Agentic Control Plane and Akuity MCP Server. This post covers the open-source Argo CD MCP Server; see the comparison below.
The Argo CD MCP Server is an open-source (Apache 2.0) Model Context Protocol (MCP) server, created by Akuity, that lets AI agents in tools such as GitHub Copilot in VS Code, Cursor, and Claude query and manage Argo CD applications in natural language. It translates requests like "sync staging-app" into Argo CD API calls and supports stdio, HTTP stream, and SSE transports.
Overview
Managing Kubernetes applications with Argo CD provides powerful GitOps capabilities. However, interacting with Argo CD often involves switching contexts – moving between your integrated development environment (IDE), such as VS Code or Cursor, the Argo CD UI, and the command line. What if you could manage and query your Argo CD applications directly from your development environment using natural language?
The rise of AI coding assistants in IDEs has significantly changed developer workflows. The Model Context Protocol (MCP) enables AI assistants to securely interact with external tools. Now, you can leverage this integration for your Argo CD workflows with the open-source Argo CD MCP Server. The project now lives in the Argo Project Labs GitHub organization as mcp-for-argocd.
What is the Argo CD MCP Server?
The Argo CD MCP Server acts as a bridge between MCP-supported AI assistants and your Argo CD instance. Instead of crafting Argo CD CLI commands or navigating the UI, you can make requests to your AI assistant, such as:
"List all applications in the production project."
"What's the sync status of the my-app application?"
"Show me the resource tree for guestbook."
"Sync the staging-app."
This server translates these natural language requests into the appropriate Argo CD API calls and brings the results back into your chat interface, streamlining your interaction with your GitOps environment.
What are the benefits of the Argo CD MCP Server?
Integrating Argo CD with AI assistants through MCP offers tangible benefits for developers and platform teams:
Boost Productivity: Get quick answers and perform Argo CD actions without leaving your IDE, minimizing context switching.
Simplify Interaction: Use natural language to query application status, view resources, or trigger syncs, making Argo CD more accessible.
Seamless IDE Workflow: Keep your focus within your primary development environment (VS Code, Cursor, etc.) for a smoother experience.
Foundation for Automation: Provides a building block for creating more sophisticated AI-driven GitOps automations and workflows in the future.
Safe Exploration: Run the server in read-only mode so the assistant can inspect applications, resources, logs, and events without changing anything.
This open-source tool aims to make interacting with Argo CD faster and more intuitive within modern development practices.
Argo CD MCP Server tools reference
The server offers a comprehensive toolkit that supports all core Argo CD features, including application management, high-level status retrieval, and access to all managed resources. This enables the AI to gather complete application context and generate meaningful, informed suggestions. Additionally, the server includes Kubernetes-specific tools for advanced use cases—for example, fetching workload logs or discovering and executing resource actions configured in Argo CD.
These tools allow for extensive observation and management of Argo CD applications and their resources directly via your AI assistant.
The full set of tools in the current release is listed below. Tools marked Write are disabled in read-only mode.
Table: Argo CD MCP Server Tools Reference
Category | Tool | What it does | Access |
|---|---|---|---|
Cluster Management | list_clusters | List all clusters registered with Argo CD | Read |
Project Management | get_appproject | Get details of a specific AppProject | Read |
Application Management | list_applications | List and filter all applications | Read |
Application Management | get_application | Get details of a specific application | Read |
Application Management | get_application_events | Get events for an application | Read |
Application Management | create_application | Create a new application | Write |
Application Management | update_application | Update an existing application | Write |
Application Management | delete_application | Delete an application | Write |
Application Management | sync_application | Trigger a sync operation on an application | Write |
Resource Management | get_application_resource_tree | Get the resource tree for an application | Read |
Resource Management | get_application_managed_resources | Get the managed resources for an application | Read |
Resource Management | get_resources | Get the manifests for an application's resources | Read |
Resource Management | get_application_workload_logs | Get logs for workloads such as Pods and Deployments | Read |
Resource Management | get_resource_events | Get events for resources managed by an application | Read |
Resource Management | get_resource_actions | Get the available actions for a resource | Read |
Resource Management | run_resource_action | Run an action on a resource | Write |
See the Argo CD MCP Server in Action
See the Argo CD MCP Server in action! This short demo shows how you can create and sync an Argo CD application using simple, natural language commands directly within your AI assistant's chat interface.
What the demo shows:
The user asks the assistant to create an Argo CD application from a Git repository.
The assistant calls create_application with the repository, path, and destination cluster and namespace.
The user asks the assistant to sync the application, and it calls sync_application.
The assistant checks the result with get_application and reports the sync and health status in the chat.
How do I install the Argo CD MCP Server?
Integrating the Argo CD MCP server into your workflow is a straightforward process, and here's how to get started.
Prerequisites
Node.js (v18+)
An Argo CD instance with API access (URL and an authentication token). See the Argo CD docs on generating an API token.
An AI assistant that supports MCP (e.g., VS Code, Cursor, Claude Desktop)
VS Code (.vscode/mcp.json)
Then start a conversation with an AI assistant in VS Code that supports MCP.
Cursor (.cursor/mcp.json)
Then start a conversation in Agent mode.
Claude Desktop (claude_desktop_config.json)
Use the same configuration as Cursor in your claude_desktop_config.json file, then point Claude Desktop at the file in its settings.
Docker and the HTTP transport
To run the server as a container:
argoprojlabs/mcp-for-argocd
The HTTP listener binds to loopback (127.0.0.1) by default. To publish a port, widen the bind address and set an inbound token:
For Kubernetes deployments with multiple replicas and no sticky sessions, run the image with http --stateless as its arguments so any pod can serve any request.
Self-signed certificates
If your Argo CD instance uses a self-signed certificate or a private CA, add "NODE_TLS_REJECT_UNAUTHORIZED": "0" to the env block. This disables TLS verification, so use it only in development or when you understand the security trade-off.
Once configured, your MCP-compatible AI assistant can start using the available Argo CD tools.
Is the Argo CD MCP Server safe to use? Read-only mode and token security
Read-only mode. Set "MCP_READ_ONLY": "true" to disable every tool that changes state: create_application, update_application, delete_application, sync_application, and run_resource_action. All read tools stay available. By default, all tools are enabled.
API token handling. The Argo CD API token is read only from environment variables or HTTP headers, never from tool-call arguments, so it never enters prompts, model context, or tool-call logs.
Network exposure. The HTTP and SSE listeners bind to loopback by default. Widening the bind requires an inbound token (MCP_AUTH_TOKEN) or an explicit --allow-unauthenticated flag; otherwise the server refuses to start.
Multiple Argo CD instances. To target several instances, each with its own token, point ARGOCD_TOKEN_REGISTRY_PATH at a JSON file that maps base URLs to tokens, for example a mounted Kubernetes secret.
Argo CD MCP Server vs. Akuity Agentic Control Plane
Akuity also offers the Agentic Control Plane, part of the Akuity Platform. Agents connect to it through the Akuity MCP Server, which covers Argo CD, Kargo, and fleet-wide delivery data, and each action runs under the identity and permissions of the user who connected the agent. The table below covers the main differences between Argo CD MCP Server and the Agentic Control Plane.
Table: Argo CD MCP Server vs. Agentic Control Plane Comparison
Argo CD MCP Server | Akuity Agentic Control Plane | |
|---|---|---|
Scope | Argo CD | Argo CD, Kargo, and fleet insights across clusters |
Identity | An Argo CD API token configured on the server | The user or API key that connected the agent |
Permissions | All tools, or read-only mode | Read-only by default; never more than the connecting user can do |
Audit | Attributed to the Argo CD account behind the token | Recorded per user and tagged as MCP actions |
Hosting and availability | Self-hosted; free and open source | Akuity Cloud; early access |
Ready to try the Argo CD MCP Server?
The Argo CD MCP Server, created by Akuity, the original creators of the Argo Project, is now available on GitHub: github.com/argoproj-labs/mcp-for-argocd. For questions, discussions, and to connect with other users and the maintainers, join the Akuity Community Discord server.
If you're interested in other helpful tools created by the minds behind the Argo Project, try out the Akuity Platform today for a supercharged Argo CD experience. Request an Akuity Platform demo for more details.
This server can become an indispensable tool for anyone practicing GitOps with Argo CD in AI-assisted development environments. We encourage you to try it out, explore its capabilities, provide feedback, and contribute to its evolution. Let's enhance the GitOps experience together!
Frequently Asked Questions
What are the prerequisites for using Argo CD MCP Server?
You need Node.js (v18 or higher) and an Argo CD instance with API access, plus an Argo CD API token and an MCP-compatible client such as VS Code, Cursor, or Claude Desktop.
Which AI assistants work with the Argo CD MCP Server?
Any MCP-compatible client. Tested configurations are provided for VS Code (.vscode/mcp.json), Cursor (.cursor/mcp.json), and Claude Desktop (claude_desktop_config.json).
Which transport protocols does it support?
It supports stdio for local IDE use, plus HTTP stream and SSE for running the server as a network service, for example in Docker or Kubernetes.
Can I stop the AI assistant from changing my applications?
Yes. Set MCP_READ_ONLY=true to disable create_application, update_application, delete_application, sync_application, and run_resource_action. All read tools stay available.
How does the server keep my Argo CD API token secure?
The token is read only from environment variables or HTTP headers, never from tool arguments, so it is not exposed to the model or written to tool-call logs.
Can one server connect to multiple Argo CD instances?
Yes. Configure a token registry: a JSON file, referenced by ARGOCD_TOKEN_REGISTRY_PATH, that maps each Argo CD base URL to its own token.
Is the Argo CD MCP Server free and open source?
Yes. It is licensed under Apache 2.0 and maintained in the Argo Project Labs GitHub organization as mcp-for-argocd. The npm package is argocd-mcp.
How is the Argo CD MCP Server different from the Akuity Agentic Control Plane?
The Argo CD MCP Server is a free, open-source server that connects an AI assistant to Argo CD using an API token you configure. The Akuity Agentic Control Plane is a hosted part of the Akuity Platform that connects agents to Argo CD, Kargo, and fleet-wide delivery data through the Akuity MCP Server, with every action tied to the connecting user, governed by your release rules, and recorded in an audit log. See the launch post for details.
