Trellis MCP

A powerful file-backed MCP (Model Context Protocol) server that implements hierarchical project management for software development teams. Organize your work with a clear structure: Projects → Epics → Features → Tasks.

Why Trellis MCP?

Trellis MCP transforms project management by providing:

• Structured Workflow: Break down complex projects into manageable, hierarchical components
• Developer-First: Built for software teams with file-based storage that integrates seamlessly with your existing tools
• AI-Native: Designed specifically for AI coding assistants like Claude, enabling intelligent task management
• Dependency Management: Support for cross-system prerequisites with cycle detection and validation
• Human-Readable: All data stored as Markdown files with YAML front-matter - no proprietary formats
• Flexible Architecture: Support both hierarchical tasks (within project structure) and standalone tasks for urgent work

Why not?

• Currently only suited to single-user projects. You probably can set the project root to a shared directory, but this is not tested.
• Not intended to replace full-fledged project management tools like Jira or Asana. Trellis MCP is focused on software development workflows and AI assistant integration.

Features

• Hierarchical project structure: Projects → Epics → Features → Tasks
• Cross-system task support: Mix hierarchical and standalone tasks with prerequisites spanning both systems
• File-backed storage: Human-readable Markdown files with YAML front-matter
• MCP server integration: JSON-RPC API for programmatic access by AI assistants
• Comprehensive validation: Cycle detection, prerequisite validation, and type safety
• Atomic operations: Task claiming, completion, and status transitions with integrity guarantees

Installation

Claude Code Configuration

Add Trellis MCP to your Claude Code MCP configuration:

# Add to Claude Code
claude mcp add task-trellis \
  -- uvx task-trellis-mcp serve

# Or specify a custom project root
claude mcp add task-trellis \
  -- uvx task-trellis-mcp --project-root /path/to/project serve

Configuration in ~/.config/claude/mcp_servers.json:

{
  "mcpServers": {
    "task-trellis": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "task-trellis-mcp",
        "serve"
      ]
    }
  }
}

VS Code with Claude Extension

Add to your VS Code settings:

{
  "claude.mcpServers": {
    "task-trellis": {
      "command": "uvx",
      "args": ["task-trellis-mcp", "serve"]
    }
  }
}

Other MCP Clients

For other MCP-compatible tools, use the command:

uvx task-trellis-mcp serve

Or with HTTP transport:

uvx task-trellis-mcp serve --http localhost:8545

Usage

MCP Tool Integration

Trellis MCP provides a comprehensive set of tools for AI assistants to manage hierarchical project structures. Once configured with your MCP client, these tools enable intelligent project planning and task management.

Core MCP Tools

• createObject - Create projects, epics, features, or tasks with validation
• getObject - Retrieve detailed object information with automatic type detection
• updateObject - Modify object properties with atomic updates
• listBacklog - Query and filter tasks across the project hierarchy
• claimNextTask - Claim tasks using priority-based, scope-based, or direct task ID selection
• completeTask - Mark tasks complete with logging and file tracking

Creating Project Hierarchies

Start by creating a project and breaking it down into manageable components:

// Create a new project
await mcp.call('createObject', {
  kind: 'project',
  title: 'E-commerce Platform Redesign',
  priority: 'high',
  projectRoot: '.',
  description: 'Comprehensive redesign of the e-commerce platform...'
});

// Create an epic within the project
await mcp.call('createObject', {
  kind: 'epic',
  title: 'User Authentication System',
  parent: 'P-ecommerce-platform-redesign',
  priority: 'high',
  projectRoot: '.'
});

// Create features within the epic
await mcp.call('createObject', {
  kind: 'feature',
  title: 'User Registration',
  parent: 'E-user-authentication-system',
  priority: 'high',
  projectRoot: '.'
});

// Create implementable tasks
await mcp.call('createObject', {
  kind: 'task',
  title: 'Create user database model',
  parent: 'F-user-registration',
  priority: 'high',
  projectRoot: '.',
  prerequisites: ['T-setup-database-schema']
});

Task Management Workflow

Use the task management tools to claim, track, and complete work:

// List available tasks
const backlog = await mcp.call('listBacklog', {
  projectRoot: '.',
  status: 'open',
  priority: 'high',
  sortByPriority: true
});

// Claim the next highest-priority task (any scope)
const claimedTask = await mcp.call('claimNextTask', {
  projectRoot: '.',
  worktree: 'feature/user-auth'
});

// Claim specific task by ID (direct claiming)
const specificTask = await mcp.call('claimNextTask', {
  projectRoot: '.',
  taskId: 'T-implement-user-auth',
  worktree: 'feature/auth-implementation'
});

// Claim specific standalone task
const standaloneTask = await mcp.call('claimNextTask', {
  projectRoot: '.',
  taskId: 'task-security-audit',
  worktree: 'hotfix/security'
});

// Claim task within specific project scope
const projectTask = await mcp.call('claimNextTask', {
  projectRoot: '.',
  scope: 'P-ecommerce-platform',
  worktree: 'feature/ecommerce'
});

// Claim task within specific epic scope
const epicTask = await mcp.call('claimNextTask', {
  projectRoot: '.',
  scope: 'E-user-authentication',
  worktree: 'feature/auth'
});

// Claim task within specific feature scope
const featureTask = await mcp.call('claimNextTask', {
  projectRoot: '.',
  scope: 'F-login-functionality',
  worktree: 'feature/login'
});

// Update task progress
await mcp.call('updateObject', {
  id: 'T-create-user-model',
  projectRoot: '.',
  yamlPatch: {
    status: 'review'
  }
});

// Complete the task with summary
await mcp.call('completeTask', {
  projectRoot: '.',
  taskId: 'T-create-user-model',
  summary: 'Implemented user model with validation and security features',
  filesChanged: ['src/models/User.js', 'tests/models/User.test.js']
});

Cross-System Prerequisites

Trellis supports complex dependency relationships across different parts of your project:

// Create a standalone urgent task that depends on hierarchy tasks
await mcp.call('createObject', {
  kind: 'task',
  title: 'Security hotfix deployment',
  projectRoot: '.',
  priority: 'high',
  prerequisites: ['T-auth-implementation', 'T-validation-update'],
  // No parent - this is a standalone task
});

Querying and Filtering

Use flexible querying to understand project status:

// Get all open tasks for a specific feature
const featureTasks = await mcp.call('listBacklog', {
  projectRoot: '.',
  scope: 'F-user-registration',
  status: 'open'
});

// Get high-priority tasks across the entire project
const urgentTasks = await mcp.call('listBacklog', {
  projectRoot: '.',
  priority: 'high',
  sortByPriority: true
});

// Get task details with prerequisites
const taskDetails = await mcp.call('getObject', {
  id: 'T-create-user-model',
  projectRoot: '.'
});

Working with AI Assistants

When using Trellis MCP with AI coding assistants, you can request natural language operations that use these tools behind the scenes:

• "Create a new project for inventory management and break it down into epics"
• "Claim the next highest priority task and implement it"
• "Claim the specific task T-implement-user-auth and work on it"
• "Work on the security audit task directly"
• "Claim a task from the user authentication epic"
• "Show me all open tasks that are ready to work on"
• "Show me tasks in the frontend development epic"
• "Complete the current task and provide a summary of what was implemented"
• "Claim a task from the login feature specifically"
• "Continue work on task T-fix-validation-bug"

Natural Language Task Claiming

AI assistants can interpret various claiming strategies:

// "Work on user authentication epic"
const authTask = await mcp.call('claimNextTask', {
  projectRoot: './planning',
  scope: 'E-user-authentication'
});

// "Focus on login functionality feature"
const loginTask = await mcp.call('claimNextTask', {
  projectRoot: './planning', 
  scope: 'F-login-functionality'
});

// "Work on the specific user registration task"
const specificTask = await mcp.call('claimNextTask', {
  projectRoot: './planning',
  taskId: 'T-user-registration-form'
});

// "Continue the security audit task"
const auditTask = await mcp.call('claimNextTask', {
  projectRoot: './planning',
  taskId: 'task-security-audit'
});

// "Claim any task from the mobile app project"
const mobileTask = await mcp.call('claimNextTask', {
  projectRoot: './planning',
  scope: 'P-mobile-app-redesign'
});

Sample Commands

For examples of how to create comprehensive AI assistant commands that leverage these MCP tools, see the sample commands directory. These examples show how to build complex workflows that combine multiple MCP tool calls for project planning and task implementation.

Direct CLI Usage

You can also use Trellis MCP directly from the command line for manual operations:

Using uv (Fast Python Package Manager)

# Install with uv
uv add task-trellis-mcp

# Or run directly without installation
uvx task-trellis-mcp serve

Development Installation

For development or to install from source:

# Clone the repository
git clone https://github.com/langadventurellc/trellis-mcp.git
cd trellis-mcp

# Install development dependencies
uv sync

# Install in editable mode
uv pip install -e .

# Initialize a new project structure
task-trellis-mcp init

# Start the MCP server
task-trellis-mcp serve

Requirements

• Python 3.12+
• Click >= 8.1
• FastMCP >= 0.7

Developer Guidelines

Quality Gate

Run all checks before committing - any failure blocks the commit:

uv run poe quality # flake8, black, pyright
uv run pytest # unit tests

Code Style

• Formatting: black and flake8 enforce code style automatically
• Type Checking: pyright ensures type safety with strict settings
• Line Limits: Functions ≤ 40 LOC, classes ≤ 200 LOC
• Import Organization: One logical concept per file
• Modern Python: Use built-in types (list, dict) over typing equivalents
• Union Types: Use str | None instead of Optional[str]

Architecture Principles

• Single Responsibility: Each module/class/function has one clear purpose
• Minimal Coupling: Components interact through clean interfaces
• High Cohesion: Related functionality grouped together
• Dependency Injection: Avoid tight coupling between components
• No Circular Dependencies: Maintain clear dependency flow

Security Requirements

• Input Validation: Validate ALL user inputs
• Parameterized Queries: Never concatenate user data into queries
• Secure Defaults: Fail closed, not open
• Least Privilege: Request minimum permissions needed
• No Hardcoded Secrets: Use environment variables and configuration

Testing Standards

• Comprehensive Coverage: Write tests alongside implementation
• Test Pyramid: Unit tests > integration tests > end-to-end tests
• Fast Feedback: Unit tests must run quickly (< 5 seconds total)
• Clear Test Names: Test names describe behavior being verified
• Isolated Tests: No dependencies between test cases

Development Workflow

Setup

# Clone repository
git clone https://github.com/langadventurellc/trellis-mcp.git
cd trellis-mcp

# Install dependencies
uv sync

# Install pre-commit hooks
uv run pre-commit install

# Install in editable mode
uv pip install -e .

Daily Development

# Format code
uv run black src/

# Lint code
uv run flake8 src/

# Type check
uv run pyright src/

# Run unit tests
uv run pytest -q

# Run all quality checks
uv run poe quality

Common Commands

Goal	Command
Install dependencies	uv sync
Start server (STDIO)	uv run task-trellis-mcp serve
Start server (HTTP)	uv run task-trellis-mcp serve --http localhost:8000
Initialize planning	uv run task-trellis-mcp init
All quality checks	uv run poe quality
Run formatter	uv run black src/
Run linter	uv run flake8 src/
Type check	uv run pyright src/
Run unit tests	uv run pytest -q

Task-Centric Development

This project uses its own task management system for development:

Working with Tasks

# Claim next available task
uv run task-trellis-mcp claim-task

# List available tasks
uv run task-trellis-mcp list tasks --status open

# Complete a task with summary
uv run task-trellis-mcp complete T-task-id \
  --summary "Implemented feature with comprehensive tests" \
  --files-changed src/module.py,tests/test_module.py

License

MIT License - See LICENSE file for details.