Agentic Tools

You have access to a complete MCP server that gives AI assistants powerful task management, agent memories, and project-specific storage. Install, run, and connect with MCP clients to manage projects, tasks, and memories across single or multiple workspaces, while keeping data isolated per project and optionally using a global workspace for cross‑project workflows.

How to use

Connect to the MCP server using an MCP client that supports stdio mode. Start the server in a project-specific workspace to manage a dedicated set of tasks and memories for that project, or use a global workspace for cross-project work. The server exposes tools for project, task, and memory management, plus AI-assisted planning and research features. Start by launching the MCP server from your terminal, then configure your client to point at the local stdio session.

Typical workflows include creating a project, adding tasks with rich metadata, breaking tasks into subtasks, tracking time and progress, and storing memories with searchable metadata. You can also run AI-assisted actions like parsing product requirements documents, obtaining intelligent task recommendations, analyzing task complexity, inferring progress, and guiding research with memory integration.

How to install

Prerequisites: you need Node.js and a basic JavaScript/TypeScript toolchain. Ensure you have npm or npx installed with Node.js.

# Quick start: run the MCP server directly without global installation
npx -y @pimzino/agentic-tools-mcp

Configuration and running in stdio mode (examples)

The MCP server can be run directly via an executable command. The following stdio configurations start the server using npx and load the MCP package. Use these in your client configuration to connect to a local, project-scoped workspace.

{
  "mcpServers": {
    "agentic_tools": {
      "command": "npx",
      "args": ["-y", "@pimzino/agentic-tools-mcp"]
    }
  }
}

Projects, tasks, and memories at a glance

Data is stored per working directory by default, inside a dedicated folder. This enables project isolation while allowing you to commit your task and memory data alongside code. Data is persisted to JSON files and can be migrated or backed up easily.

Security and data practices

All inputs are validated and operations are designed to be atomic to prevent partial data writes. Destructive actions require explicit confirmation. When using a global workspace, understand that data is shared across projects in that workspace.

Troubleshooting

If you encounter issues starting the server, verify that Node.js is installed and reachable from your shell, and confirm that your working directory exists and is writable. Check for clear error messages that point to missing files, permission problems, or invalid configuration.

Data storage structure

Project-specific data is stored under your working directory in a folder named .agentic-tools-mcp/. Tasks reside in .agentic-tools-mcp/tasks/ and memories in .agentic-tools-mcp/memories/. This structure supports easy backup and version control alongside your project files.

Agent memories and task tools overview

The server provides a rich set of tools to manage projects, tasks, subtasks, and memories, plus AI-assisted capabilities for product requirements parsing, task recommendations, and memory-based research.

With Claude Desktop or other clients

For Claude Desktop, you can use a global directory by adding the appropriate flag to the command, which switches storage to a centralized global workspace. When using a global workspace, the workingDirectory parameter is ignored and all data is stored in the global location.

With the VS Code extension

Pair the MCP server with the Agentic Tools MCP Companion extension for a GUI interface that visualizes tasks, priorities, statuses, and memories. Use both together to gain real-time synchronization and a seamless workflow between code and AI planning.

Available tools

list_projects

View all projects in a working directory

create_project

Create a new project in a working directory

get_project

Get detailed project information

update_project

Edit project name or description

delete_project

Delete a project and all associated data

list_tasks

View tasks in a hierarchical tree with unlimited depth

create_task

Create a task at any level with an optional parentId to nest under a parent task

get_task

Get detailed task information including hierarchy relationships

update_task

Edit task metadata or move between hierarchy levels using parentId

delete_task

Delete a task and all its subtasks recursively

move_task

Move a task within the hierarchy to a new parent or level

migrate_subtasks

Automatically migrate legacy subtasks to the unified Task model

parse_prd

Parse Product Requirements Documents and generate structured tasks

get_next_task_recommendation

Get intelligent task recommendations based on dependencies, priorities, and complexity

analyze_task_complexity

Analyze task complexity and suggest breaking down overly complex tasks

infer_task_progress

Infer progress by analyzing codebase and implementation evidence

research_task

Guide AI agents to perform web research with memory integration

generate_research_queries

Generate targeted web search queries for task research

list_subtasks

View child tasks (legacy compatibility, now mapped to unified Task model)

create_subtask

Create a child task under a parent task (legacy compatibility)

get_subtask

Get information for a legacy subtask (mapped to unified Task model)

update_subtask

Edit a legacy subtask (mapped to unified Task model)

delete_subtask

Delete a legacy subtask (mapped to unified Task model)

create_memory

Store new memories with title and detailed content

search_memories

Find memories using intelligent multi-field search with relevance scoring

get_memory

Get detailed memory information

list_memories

List memories with optional filtering

update_memory

Edit memory title, content, metadata, or categorization

delete_memory

Delete a memory