Task Trellis MCP

Project planning and task management built specifically for AI agents

Task Trellis is an MCP server for project planning and task management built specifically for AI agents. It helps by breaking down complex projects and tracking their progress with built-in task management, complete with progress tracking, dependency management, and workflow automation. By default, all data is stored locally in Markdown files.

Primarily built as a much better alternative to managing markdown checklists. Task Trellis will make it easier to define requirements, specifications, and tasks in a structured way that the agents can actually use directly.

Full documentation is available in the docs folder.

Table of Contents

• At a Glance

• Why Task Trellis?

• Core Benefits

• Usage

• Installation and Configuration

• Configuration

• Available Tools

• Troubleshooting

Related MCP server: TaskMateAI

At a Glance

Why Task Trellis?

Without Task Trellis

• AI agents lose track of complex, multi-step projects

• Agents spin out of control with no clear task structure

• Tasks are often too large or vague, leading to confusion

• No way to manage dependencies or prerequisites

• No visibility into what's been completed vs. what's pending

• Tasks get forgotten, duplicated, or done out of order

• Zero coordination between multiple AI sessions

• Complex projects become chaotic and overwhelming

With Task Trellis

• Structured Breakdown: Automatically organize projects into hierarchical tasks (depending on the size of the effort required)

  • Project → Epic → Feature → Task

  • See Picking a Parent Issue Type

• Smart Dependencies: Prevent tasks from starting until prerequisites are complete

• Progress Tracking: Real-time visibility into what's done, in-progress, and pending

• Session Continuity: Pick up exactly where you left off across AI conversations

• Workflow Management: Built-in task claiming, completion, and validation workflows

• File Change Tracking: Automatic documentation of what files were modified for each task

• Learn from History: AI agents can reference past work to inform future tasks

Core Benefits

Focused Execution: AI agents work on one clearly-defined task at a time
Progress Visibility: Always know project status and what's next
Dependency Management: Automatic task ordering based on prerequisites
Audit Trail: Complete history of all work completed and changes made
Multi-Session Support: Seamlessly collaborate across different AI conversations
Productivity Boost: Reduce context switching and eliminate forgotten tasks

Usage

See full documentation at Task Trellis MCP Documentation

Basic Workflow

1. Create Tasks

   • Determine your starting point based on the expected size of your project

     • Project - For sprawling initiatives with many moving parts

     • Epic - For large feature groupings

     • Feature - For specific functionality

     • Task - For individual work items

2. Claim & Work on Tasks

   • AI agent claims next available task

     • Excludes tasks that have incomplete prerequisites

     • Grabs the next highest priority available task

     • Mark a task as draft if you don't want it to be worked on yet - it won't be claimed when the tool looks for the next available task

   • Works on the specific task requirements

   • Marks task complete with file changes documented

     • Automatically tracks which files were modified

     • Logs summary of changes made

     • Work done in the future could reference this to better understand the current state of the project

3. Track Progress

   • View completed vs. pending work

   • See dependency relationships

   • Monitor overall project health

Installation and Configuration

See installation instructions.

Configuration

CLI Flags and Environment Variables

The MCP server resolves the project directory in this order: --projectDir flag → $TRELLIS_PROJECT_DIR → current working directory. Since Claude Code launches MCP servers with the workspace as CWD, no flag is normally needed.

Shared Data Directory Layout

Data is stored in ~/.trellis/ (shared across all sessions), not inside the repo directory.

~/.trellis/
  projects/
    <12-char-key>/        ← sha1(gitOriginUrl or absolutePath).slice(0,12)
      p/ e/ f/ t/         ← issues (unchanged internal layout)
      meta.json           ← { "label": "<gitOriginUrl or absolutePath>" }

Browser UI

When the first Claude Code session starts the MCP server, it binds http://127.0.0.1:3717 and logs:

Task Trellis UI: http://127.0.0.1:3717

• Subsequent sessions detect the port is taken and run STDIO-only.

• The UI shows all projects under ~/.trellis/projects/ and supports creating, editing, and deleting issues directly in the browser.

• When the leader session exits, the port is released automatically.

Project Tree Navigation

The project tree view includes several features for navigating and filtering issues:

• Expand/Collapse: Click the chevron next to an issue to expand or collapse its children. Open/closed state is automatically saved per project.

• Hide Completed: Click the filter icon (top toolbar) to toggle hiding completed (done/wont-do) issues. This preference is saved per project.

• Search: Use the search box to filter issues by title or content. Search results are displayed flat and ignore the hide-completed preference.

• Issue Details: Click any issue row to view and edit its full details in the side panel.

Breaking Changes

Breaking change: --projectRootFolder has been removed. Use --projectDir instead.

Breaking change: The activate MCP tool has been removed. The project directory defaults to the current working directory; override with --projectDir or $TRELLIS_PROJECT_DIR.

Note: Data previously stored in <repo>/.trellis/ is not migrated automatically. Move or recreate your data under ~/.trellis/.

Available Tools

Core Issue Management

• create_issue - Create projects, epics, features, or tasks with hierarchical relationships

• update_issue - Modify issue properties, status, priority, prerequisites, or labels

• get_issue - Retrieve detailed issue information with history and relationships

• list_issues - Query and filter issues by type, status, priority, or scope (returns issue summaries)

• delete_issue - Remove issues (with dependency validation)

• replace_issue_body_regex - Make targeted body content edits using regex patterns

Task Workflow Management

• claim_task - Claim available tasks for execution with automatic priority ordering

• complete_task - Mark tasks complete with file change documentation

• get_next_available_issue - Use this tool to find the next available issue that's ready to work on.

• append_issue_log - Add progress notes and status updates to task history (occurs automatically on task completion)

• append_modified_files - Record files modified during task execution with change descriptions (occurs automatically on task completion)

Attachment Management

• add_attachment - Copy a file into the managed attachments folder for an issue (errors if the issue or source file does not exist, or a file with the same name already exists)

• remove_attachment - Delete a named file from an issue's attachments folder (errors if the issue or file does not exist)

Attachments are returned as a list of filenames when calling get_issue and are linked in the browser UI detail view.

Project Files

• write_project_file - Write a UTF-8 text file to the project's files directory. Overwrites by default; pass failIfExists: true to error on collision instead.

• read_project_file - Read a UTF-8 text file from the project's files directory.

• list_project_files - List filenames in the project's files directory (returns [] if the directory does not yet exist).

• delete_project_file - Delete a named file from the project's files directory.

Project files live at ~/.trellis/projects/<key>/files/<filename> and are scoped to the project — not tied to any individual issue. Filenames must not contain path separators or .. (path-traversal protection); v1 supports a flat namespace of UTF-8 text files.

Browser UI

• get_ui_info - Returns the URL and port of the Task Trellis browser UI; use when the user asks about the UI or wants to view issues in a browser

System Management

• prune_closed - Clean up old completed/cancelled issues for maintenance

Troubleshooting

Common Issues

Configuration issues:

• Validate JSON syntax in MCP client configuration

• Ensure absolute paths are used for --projectDir

• Restart your MCP client after configuration changes

Getting Help

• Issues: Report bugs or feature requests

• Documentation: Check this README and docs

License

GPL-3.0-only - see LICENSE file for details.