OmniPlan MCP

An MCP (Model Context Protocol) server and CLI for reading and writing OmniPlan (.oplx) and Microsoft Project (.mpp) schedule files.

macOS only — AppleScript bridge required for .mpp and all write operations.

Installation

pip install omniplan-mcp

Or install from source in editable mode:

cd /Users/cygnus/work/github/omniplan-mcp
pip install -e .

Related MCP server: projectlibre-mcp

Usage

CLI mode (human-friendly)

Read and display a schedule:

# Read a complete schedule
omniplan-mcp read schedule.oplx

# Show a high-level summary
omniplan-mcp summary schedule.oplx

# Search for tasks by name
omniplan-mcp search schedule.oplx "design"

# List all tasks (table or tree view)
omniplan-mcp tasks schedule.oplx
omniplan-mcp tasks schedule.oplx --tree

# List resources or dependencies
omniplan-mcp resources schedule.oplx
omniplan-mcp dependencies schedule.oplx

Modify a schedule (requires the document to be open in OmniPlan):

# Find a task's ID by name
omniplan-mcp lookup "Task Name"

# Mark a task as complete
omniplan-mcp set-done 258
omniplan-mcp set-done 258 --subtree

# Add/remove dependencies
omniplan-mcp add-dep 260 258
omniplan-mcp rm-dep 260 258

# Set duration (seconds, days, hours, or minutes)
omniplan-mcp set-duration 258 3d
omniplan-mcp set-duration 258 28800

# Add a new child task
omniplan-mcp add-task 258 "Subtask" 2d

# Rename, delete, clear constraints
omniplan-mcp rename 258 "New Name"
omniplan-mcp delete 258
omniplan-mcp clear-constraint 258

# Save the document
omniplan-mcp save

# JSON output (read commands)
omniplan-mcp read schedule.oplx --json
omniplan-mcp summary schedule.oplx --json

Get help:

omniplan-mcp --help
omniplan-mcp read --help

MCP server mode (for AI tools)

Start the MCP server in stdio mode:

omniplan-mcp serve

Configure your MCP host (e.g., Claude Code) to use it:

{
  "mcpServers": {
    "omniplan": {
      "command": "omniplan-mcp",
      "args": ["serve"]
    }
  }
}

For source installs managed by uv, keep the serve subcommand in the configured arguments:

{
  "mcpServers": {
    "omniplan": {
      "command": "uv",
      "args": [
        "run",
        "--directory",
        "/path/to/omniplan-mcp",
        "omniplan-mcp",
        "serve"
      ],
      "env": {}
    }
  }
}

If your MCP host reports that the server failed to connect, check that the configured command starts omniplan-mcp serve. Running omniplan-mcp without serve starts the CLI help path and exits instead of staying open as a stdio MCP server.

Architecture

src/omniplan_mcp/
├── __init__.py      # Version (__version__ = "0.5.0")
├── __main__.py      # CLI entry point: delegates to cli.py
├── cli.py           # CLI subcommands (click)
├── server.py        # MCP server: tool definitions + output formatters
└── parser.py        # Two parsing paths + write operations (AppleScript bridge)
tests/
└── test_parser.py   # Unit tests with in-memory .oplx ZIPs

Dual-mode design

The package provides two interfaces from the same codebase:

1. CLI mode (omniplan-mcp read ..., omniplan-mcp set-done ...) — human-friendly terminal output

2. MCP server mode (omniplan-mcp serve) — JSON-RPC over stdio for AI tools

Both share the same parser (parser.py) and AppleScript bridge.

Key design decisions

1. Dual parser architecture: .mpp files open OmniPlan and read via AppleScript's in-memory object model. .oplx files parse XML directly (no OmniPlan needed). Both return identical 6-tuples: (projects, resources, tasks, violations, assignments, dependencies).

2. Two ID systems: XML (.oplx) uses string IDs like "t258". AppleScript uses sequential integers starting at 1. All write operations strip the t prefix automatically.

3. Write operations work on the open OmniPlan document: Tools like add_dependency, set_task_duration, clear_constraint_date generate AppleScript that targets document 1 of application "OmniPlan". The document must be open.

4. Read tools work from file (.oplx or .mpp): read, summary, search, tasks, resources, dependencies parse the file on disk.

5. list_dependencies reads from the baseline scenario, not the editing scenario.

Omni Automation JavaScript vs JXA

evaluate_omniplan_script uses Omni Automation JavaScript (not JXA/AppleScript JS). See the full explanation in the MCP server documentation.

Percent-complete computation

.oplx files store completion as effort-done / effort ratio. Group tasks compute completion bottom-up from children. Task status is computed: 100% → "finished", else compare end date to today.

Development

# Install in editable mode
pip install -e .

# Run tests
python -m pytest tests/ -v

# Build distribution
python -m build

Testing

Tests use a hand-crafted .oplx ZIP in memory (no real files). Add new test functions in tests/test_parser.py with inline XML constants. Tests cover: parsing, resource filtering, tree building, string parent IDs, percent-complete from effort, outline_depth, task_status, and Actual.xml preference.