All Products
Search
Document Center

Alibaba Cloud DevOps:Alibaba Cloud DevOps MCP Server

Last Updated:Aug 27, 2026

The Alibaba Cloud DevOps MCP Server (alibabacloud-devops-mcp-server) enables AI assistants to interact with the Alibaba Cloud DevOps platform. AI assistants can read work item content in projects, understand requirements, automatically write code, and submit merge requests. R&D teams can use this tool to assist with code reviews, optimize task management, and reduce repetitive operations, allowing them to focus on innovation and product delivery. The Alibaba Cloud DevOps MCP Server is an open-source project. Source code: https://github.com/aliyun/alibabacloud-devops-mcp-server.

Features

After connecting, you can use natural language to instruct AI assistants (such as Cursor, Claude, and Tongyi Lingma) to check work items, read code, submit merge requests, and run pipelines. The following capabilities are covered:

  • Organization management: Organization list, organization information, department information, organization roles, member information, etc.

  • Code management (Codeup): Code repositories, branches, merge requests, file trees, etc.

  • Project collaboration (Projex): Projects, work items, work item fields, comments, work hour management, etc.

  • Pipeline management (Flow): Pipelines, tasks, resources, tags, deployment management, etc.

  • Artifact repository (Packages): Artifact repositories, artifact list, etc.

  • Application delivery (AppStack): Deployment orders, applications, application tags, variable group management, etc.

  • Test management (Testhub): Test cases, test case directories, test plans, test results, etc.

Prerequisites

  • Obtain a personal access token, and grant the token read and write permissions for all APIs under Organization Management, Project Collaboration, Code Management, Pipeline, Artifact Repository, Application Delivery, and Test Management. Set a sufficiently long expiration period for the token.

  • OAuth 2.0 authorization: If your MCP client supports OAuth, you do not need to create a token in advance. After configuring the remote hosting URL, the client will automatically guide you to log in and authorize through a browser. For details, see "Advanced configuration > Authentication methods > OAuth 2.0 Authorization" below.

  • (stdio mode only) Node.js version >= 20.0.0.

Access methods

The Alibaba Cloud DevOps MCP Server supports two access methods: Remote Hosting (Streamable HTTP) and Local (stdio). Remote Hosting requires no dependency installation — simply configure and use. Local mode requires a Node.js environment and uses npx to pull and start the service. The service addresses below use the central station as an example. For Region stations, replace the address with your organization-specific domain + /ai/mcp. For details, see "Region configuration (Dedicated Edition)" below.

Remote hosting (Streamable HTTP)

Replace <YOUR_TOKEN> with your personal access token:

{
  "mcpServers": {
    "yunxiao": {
      "url": "https://openapi-rdc.aliyuncs.com/ai/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_TOKEN>"
      }
    }
  }
}

If your client does not support custom headers, you can use a Query String to pass the token instead: https://openapi-rdc.aliyuncs.com/ai/mcp?yunxiao_access_token=<YOUR_TOKEN>.

If your client supports OAuth, you can skip the token and use browser-based authorization instead. For details, see "Advanced configuration > Authentication methods > OAuth 2.0 Authorization" below.

Local (stdio)

{
  "mcpServers": {
    "yunxiao": {
      "command": "npx",
      "args": ["-y", "alibabacloud-devops-mcp-server"],
      "env": {
        "YUNXIAO_ACCESS_TOKEN": "<YOUR_TOKEN>"
      }
    }
  }
}

Usage

After successfully connecting, use natural language to instruct the AI. For example:

  • "List the organizations I belong to"

  • "Show me the latest 10 commits on the master branch of the repo-name repository in organization xxx"

  • "Create a merge request from the current branch to master with the title ..."

  • "Show me the unfinished work items assigned to me in project yyy"

  • "Add a comment to work item #12345: integration testing completed"

  • "Which pipeline failed most recently? Pull up the failure logs"

The AI will automatically select the appropriate tools and execute them with your credentials.

Advanced configuration

Toolset filtering

The Alibaba Cloud DevOps MCP Server provides 194 tools covering 8 categories. Having too many tools can consume significant model context. If you only need certain categories, you can specify toolsets so the server only delivers the corresponding tools (basic tools are always included).

  • Remote Hosting mode: Append ?toolsets=... to the URL, or add the request header X-Devops-Toolsets: ....

  • stdio mode: Add the command-line argument --toolsets=..., or set the environment variable DEVOPS_TOOLSETS=....

Available values (comma-separated): code-management, organization-management, project-management, pipeline-management, packages-management, application-delivery, test-management. If not specified, all tools are enabled by default.

Remote access example:

{
  "mcpServers": {
    "yunxiao": {
      "url": "https://openapi-rdc.aliyuncs.com/ai/mcp?toolsets=code-management,project-management",
      "headers": { "Authorization": "Bearer <YOUR_TOKEN>" }
    }
  }
}

Authentication methods

The Alibaba Cloud DevOps MCP Server supports two types of authentication: manually passing a Personal Access Token (PAT), or using OAuth 2.0 authorization for supported clients to connect without a token.

Using a Personal Access Token (PAT)

The following three methods are available for passing a personal access token, listed by recommended priority:

Method

Usage

Description

Authorization header

Authorization: Bearer <TOKEN>

Preferred method, secure and universal

X-Yunxiao-Token header

X-Yunxiao-Token: <TOKEN>

Alternative method

Query String

?yunxiao_access_token=<TOKEN>

Not recommended. Appears in access logs. Use only as a temporary workaround when the client does not support custom request headers.

In stdio mode, pass the token through the YUNXIAO_ACCESS_TOKEN environment variable. The above configurations are not needed.

OAuth 2.0 Authorization

For MCP clients that support OAuth, such as Qoder/QoderCLI and Claude Code, you do not need to create a personal access token in advance. In Remote Hosting mode, after entering the service URL, the client will automatically initiate the OAuth 2.0 authorization code + PKCE flow, opening a browser to guide you through login and authorization, then automatically obtaining and refreshing tokens. This applies to Remote Hosting (Streamable HTTP) access.

When configuring, enter only the service URL without the Authorization header:

{
  "mcpServers": {
    "yunxiao": {
      "url": "https://openapi-rdc.aliyuncs.com/ai/mcp"
    }
  }
}
  • Service URL: Central station https://openapi-rdc.aliyuncs.com/ai/mcp; Region station https://<organization-specific-domain>/ai/mcp.

  • Endpoint discovery, client registration, and PKCE are all handled automatically by the client. You do not need to manually apply for a client_id or configure a client_secret (the platform uses public clients).

  • Access tokens issued after authorization are valid for 24 hours, and refresh tokens for 90 days. The client automatically refreshes tokens. Re-authorization is required after the refresh token expires.

Region configuration (Dedicated Edition)

If you are using an Alibaba Cloud DevOps Region station, each Region independently provides an officially hosted MCP endpoint. The usage is the same as the central station — simply replace the service address with your organization-specific domain.

Method 1: Remote Hosting (Streamable HTTP, recommended, no installation required)

  • Service address: https://<organization-specific-domain>/ai/mcp, for example https://your-org.devops.aliyuncs.com/ai/mcp (your regular Alibaba Cloud DevOps Region domain with /ai/mcp appended)

  • Authentication: Authorization: Bearer <YOUR_TOKEN> (or X-Yunxiao-Token: <YOUR_TOKEN>). If your client supports OAuth, you can skip the token and use browser-based authorization instead. For details, see "Advanced configuration > Authentication methods > OAuth 2.0 Authorization" above.

{
  "mcpServers": {
    "yunxiao": {
      "url": "https://your-org.devops.aliyuncs.com/ai/mcp",
      "headers": { "Authorization": "Bearer <YOUR_TOKEN>" }
    }
  }
}

Method 2: Local (stdio)

Use the YUNXIAO_API_BASE_URL environment variable to point to your Region domain:

{
  "mcpServers": {
    "yunxiao": {
      "command": "npx",
      "args": ["-y", "alibabacloud-devops-mcp-server"],
      "env": {
        "YUNXIAO_ACCESS_TOKEN": "<YOUR_TOKEN>",
        "YUNXIAO_API_BASE_URL": "https://your-org.devops.aliyuncs.com"
      }
    }
  }
}

Self-deployment and developer documentation

To deploy locally or in a private environment (stdio / SSE / Streamable HTTP), run with Docker, or learn about the project's implementation details, see the full documentation on GitHub:

Related links

  • GitHub repository

  • For questions, join the Alibaba Cloud DevOps AI assistant discussion group on DingTalk (group number: 134400004101).