Port IO MCP Server

The Port MCP Server bridges your Port.io developer portal with AI tools like Claude, Cursor, VS Code, and GitHub Copilot. It exposes your software catalog, blueprints, entities, scorecards, and Port AI agents through the Model Context Protocol, letting LLMs answer questions like "who owns this service?", "what production services are failing their security scorecard?", or "create a new microservice blueprint with these properties."

Port offers two implementations. The actively maintained option is the Remote MCP Server, a cloud-hosted endpoint at https://mcp.port.io/v1 (EU) or https://mcp.us.port.io/v1 (US) that uses OAuth2 for authentication and requires no local installation. The original port-labs/port-mcp-server repo (a Python/Docker server using client credentials) is now deprecated in favor of the remote server, though it still works for self-hosted setups.

Notable capabilities include reading and writing blueprints and entities, managing scorecards and quality gates, invoking Port AI agents with custom prompts, and respecting Port's RBAC permissions so the AI cannot do anything the authenticated user could not. A x-read-only-mode header lets teams restrict the server to safe read-only queries.

Tools

Prerequisites

• A Port.io account (sign up at port.io)
• For the remote server: an MCP client that supports remote MCP (Claude Desktop, Cursor, VS Code with Copilot, etc.)
• For token-based auth: a Port Client ID and Client Secret from Settings > Credentials in the Port dashboard

Option A: Remote MCP Server (recommended)

Port hosts the MCP server at a regional URL. Authentication happens via browser-based OAuth.

Regional endpoints:

• EU: https://mcp.port.io/v1
• US: https://mcp.us.port.io/v1

Cursor config

{
  "mcpServers": {
    "port-eu": {
      "url": "https://mcp.port.io/v1",
      "headers": {
        "x-read-only-mode": "0"
      }
    }
  }
}

VS Code config (uses mcp-remote bridge)

{
  "mcpServers": {
    "port-vscode-eu": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.port.io/v1"]
    }
  }
}

Claude Desktop

Install the Port connector from Claude's connector library and enter the regional URL. Sessions last roughly 30 days with a 15-day idle timeout.

Set x-read-only-mode to 1 to restrict the server to read-only operations.

Option B: Self-hosted (deprecated)

The legacy port-labs/port-mcp-server repo runs locally via Docker or uvx.

{
  "mcpServers": {
    "port": {
      "command": "docker",
      "args": [
        "run", "-i", "--rm",
        "-e", "PORT_CLIENT_ID",
        "-e", "PORT_CLIENT_SECRET",
        "-e", "PORT_REGION",
        "ghcr.io/port-labs/port-mcp-server:latest"
      ],
      "env": {
        "PORT_CLIENT_ID": "<your-client-id>",
        "PORT_CLIENT_SECRET": "<your-client-secret>",
        "PORT_REGION": "EU"
      }
    }
  }
}

Port recommends migrating off this option to the remote server.

• Catalog Q&A in your IDE: Ask Cursor or Claude "who owns the payments-api service?" or "list all production services running Node 18" without leaving the editor.
• Scorecard and compliance reporting: Have an agent pull current scorecard scores for every service and summarize which ones are failing security or maturity rules.
• Bootstrap new services: Ask the AI to create a new entity from a microservice blueprint with the correct owner, repo, and environment fields populated.
• Build the portal itself: Use natural language to draft blueprints, scorecards, and relations, then push them to Port without writing YAML or clicking through the UI.
• Run Port AI agents in workflows: Invoke pre-configured Port AI agents from inside Claude or Copilot to trigger guided actions and operational procedures.

• "List every service in Port that is failing the production-readiness scorecard."
• "Who is the on-call owner for the checkout-service entity?"
• "Create a new blueprint called lambda-function with properties for runtime, region, and owner team."
• "Add a slack-channel property to the service blueprint and backfill it for all entities owned by the platform team."
• "Invoke the incident-response AI agent with the prompt: summarize this week's SEV-2 incidents."

• Official server maintained by Port Labs with a hosted remote option, so no local infrastructure is needed.
• Full read and write coverage of Port's core primitives: blueprints, entities, scorecards, and AI agents.
• OAuth2 authentication respects each user's Port RBAC permissions, plus an x-read-only-mode header for safer setups.
• Works across Claude Desktop, Cursor, VS Code, and any client that supports remote MCP or mcp-remote.

• Requires a Port.io account and is only useful if your organization already uses Port as its developer portal.
• The original self-hosted port-mcp-server repo is deprecated, so teams running it need to migrate to the remote server.
• Tool surface is focused on Port primitives; integrations with underlying systems (Git, K8s, cloud) still need to be configured inside Port itself.

• Backstage MCP servers (community implementations) for teams using Backstage as their developer portal instead of Port.
• GitHub MCP Server for service ownership and repo metadata when you do not need a full catalog layer.
• Cortex and OpsLevel developer portals, which offer their own API integrations though no first-party MCP server.