MCPlanManager

MCPlanManager is an MCP server that coordinates long-running AI Agent tasks. It supports the MCP standard and can be deployed via lightweight uvx-based execution or isolated Docker containers. It enables you to manage task plans, track progress, and visualize dependencies, making it ideal for orchestrating complex AI workflows.

How to use

You connect your MCP client to MCPlanManager to manage a task plan end-to-end. Use theHTTP mode for remote operation via SSE, or run a local instance with uvx for quick experimentation. Your client can initialize plans, start tasks, monitor status, export or restore plans, and visualize dependencies as you orchestrate multi-step AI activities.

Choose one of the deployment methods below and configure your MCP client to point at the appropriate endpoint. The server exposes actions such as creating a plan, adding tasks, starting the next task, marking tasks complete or failed, and exporting or loading full plan state.

How to install

# Prerequisites: ensure you have Python and uvx utilities available on your system
# Install via uvx for a lightweight start
curl -LsSf https://astral.sh/uv/install.sh | sh

# Alternatively, install Docker and pull the latest image
docker pull donway19/mcplanmanager:latest

Deployment options

Two deployment options are provided. Use the one that matches your environment and maintenance preferences.

Option 1: uvx (lightweight and fast startup) Use a single command to start the MCP server and let it manage dependencies.

Option 2: Docker (production-ready and isolated) Run the server inside a container for consistency across environments.

Configure your MCP client

Configure your MCP client to communicate with the server using either SSE (Docker) or standard I/O (uvx). The following configurations illustrate how to point your client to the running service.

Docker deployment example

{
  "mcpServers": {
    "mcplanmanager-docker": {
      "transport": "sse",
      "url": "http://localhost:8080/sse"
    }
  }
}

uvx deployment example

{
  "mcpServers": {
    "mcplanmanager": {
      "command": "uvx",
      "args": ["mcplanmanager"]
    }
  }
}

Notes on cloud deployment

If you deploy on a cloud server, replace localhost with the server’s public IP or domain when configuring the client.

Running tests

Use the project’s test suite to validate the deployment in either SSE (Docker) mode or uvx mode. Ensure the service is running before executing tests.

Available tools

initializePlan

Initialize a new task plan with a defined set of tasks and initial state.

loadPlan

Load a complete plan object and replace the current plan state.

dumpPlan

Export the current plan data as a dictionary for persistence or restoration.

getCurrentTask

Retrieve the task currently in progress.

startNextTask

Begin execution of the next available task in the sequence.

completeTask

Mark the active task as completed.

failTask

Mark the active task as failed.

skipTask

Skip a specified task in the plan.

addTask

Append a new task to the current plan.

getTaskList

Fetch tasks with optional status filtering.

getExecutableTaskList

List tasks that are currently executable.

getPlanStatus

Query the overall status of the plan.

editDependencies

Modify dependencies between tasks.

visualizeDependencies

Generate a visual representation of dependencies in ascii, tree, or mermaid formats.

generateContextPrompt

Create a context prompt to guide AI agents within the plan.