AzureDevOpsMcpServer

AzureDevOpsMcpServer is a multi-agent MCP server that exposes three families of tools (manag er, infra, and QA) to interact with Azure DevOps. The Manager Agent focuses on project, work item, iterations, and wiki interactions; the Infra Agent covers builds, pipelines, git repositories, and pull requests; the QA Agent provides test plan, test suite, and test run management. Copilot Studio users should expect flat parameter shapes and JSON strings for any complex payload to avoid nested object issues. Tools are designed to return compact, actionable results or a serialized JSON string when a deeply nested structure would otherwise be needed. To start a server instance, run it via Docker and point your requests to the exposed API endpoint. Authentication and organization details can be provided via environment variables or request-scoped credentials as supported by the deployment.

Prerequisites:

• Docker installed and running on your host
• Access to the Azure DevOps organization and a personal access token (PAT) with appropriate scopes
• Basic familiarity with MCP tooling and how to invoke agent tools

Step 1: Pull and run the AzureDevOpsMCP server image

# Example: run the MCP server image in detached mode and map port 5000
docker run -d --name azuredevops-mcp \
  -p 5000:5000 \
  -e ASPNETCORE_ENVIRONMENT=Production \
  -e AZURE_DEVOPS_ORG=your-organization-name \
  -e AZURE_DEVOPS_PAT=your-personal-access-token \
  -e MCP_BIND_ADDRESS=0.0.0.0 \
  -e MCP_BIND_PORT=5000 \
  avanade-group-supplychaincloud-azuredevopsmcpserver

Step 2: Verify the server is running

docker ps -f name=azuredevops-mcp

Step 3: Configure your client to talk to the MCP server endpoint

• Host: http(s)://<host-IP-or-domain>:5000
• Use your preferred MCP client tooling to invoke the agents (Manager, Infra, QA) and their tools.

Step 4: Optional - Run in a production-ready orchestrator

• Use docker-compose or Kubernetes to manage the container lifecycle and environment separation.

Notes:

• If you prefer to run locally without Docker, you can adapt the config to a native runtime if a compatible image or binary is provided by your team.
• Ensure the PAT is kept secret and rotated regularly; never commit credentials into source control.

Tips and considerations:

• Copilot Studio compatibility: the server avoids deeply nested object schemas. If you need to pass complex data, serialize it to a single JSON string and supply it via a parameter named <payloadJson> (e.g., updatesJson, daysOffJson).
• Outputs are flattened when possible to ease consumption in low-coverage environments or by Copilot Studio.
• For Azure DevOps interactions, ensure the PAT has at least read/write permissions for projects, work items, builds, pipelines, and repos as needed by the respective tools.
• If you encounter API shape changes from Copilot Studio, consider providing a minimal flattened response or a JSON string payload to retain compatibility.
• Environment variables can be extended to support more granular control (e.g., timeouts, rate limits, logging level).