PackageFlow

PackageFlow exposes an MCP server that lets AI tools control and query your local DevOps environment. Through MCP, AI assistants can list projects, inspect npm scripts, run scripts, manage workflows, and trigger deployments across multiple repositories in a secure, permissioned way. The MCP integration enables actions such as list_projects, get_project, run_npm_script, run_workflow, and other domain-specific actions that PackageFlow supports. Once MCP is enabled, your AI tools can call these actions to orchestrate tasks without requiring manual command-line interactions.

To use MCP with PackageFlow, start the MCP server (as described in installation) and configure your AI tool to connect via the MCP transport (stdio in PackageFlow’s case). In your AI tool, copy the generated MCP config and paste it into the tool’s MCP settings. After connection, you can issue natural-language style prompts like: “List all projects and show dev scripts,” “Run tests for the frontend repo and summarize failures,” or “Switch to the main branch and start the dev server for project-x.” The tool will translate these prompts into the appropriate MCP actions and present results back to you in a readable format.

Prerequisites:

• Docker installed and running on your machine (for the example config using docker).
• Access to the internet for pulling the MCP server image.
• Basic familiarity with your AI tool’s MCP configuration.

1. Install Docker (if not already installed).

2. Pull or build the MCP server image (example uses packageflow-mcp-image):

   docker pull packageflow-mcp-image

3. Run the MCP server:

   docker run -i packageflow-mcp-image

   Note: If your environment requires a custom image tag or registry, adjust the command accordingly.

4. Verify the MCP server is listening on the expected transport (stdio for many local setups). If needed, expose ports or configure the transport in your AI tool to connect to the MCP server.

5. Retrieve or generate the MCP configuration snippet from PackageFlow Settings > MCP > MCP Integration and copy it into your AI tool’s MCP configuration.

6. Test a simple action such as listing projects or getting a project’s scripts to confirm the integration is working.

Tips:

• MCP is local-first and permissioned. Use the provided env variables to control logging and transport behavior.
• If you encounter connection issues, verify Docker has sufficient permissions to run containers and that the MCP image is accessible.
• For security, consider enabling restricted permissions for the AI tool and enabling logging to audit actions taken via MCP.
• The MCP integration supports actions like list_projects, get_project, run_npm_script, and run_workflow. If a needed action isn’t exposed by default, consult PackageFlow documentation for additional MCP actions or custom adapters.
• Ensure your environment variables are set according to your security and networking requirements, especially if operating behind proxies or in restricted networks.