Model Context Protocol (MCP) server for OmniFocus on macOS, with Rust, Python, and TypeScript implementations
OmniFocus MCP
MCP server that gives AI assistants full control over OmniFocus on macOS.
45 tools, 3 resources, and 4 prompts covering tasks, projects, tags, folders, perspectives, forecast, notifications, and review workflows — all through the Model Context Protocol.
This project is not affiliated with, endorsed by, or associated with The Omni Group or OmniFocus. OmniFocus is a trademark of The Omni Group. This is an independent, non-commercial open-source project.
Quick Start
Install via Homebrew (if you don't have Homebrew, see the Homebrew installation guide):
brew tap vitalyrodnenko/omnifocus-mcp
brew install omnifocus-mcp
Then add to your MCP client config (Claude Desktop, Cursor, etc.):
{
"mcpServers": {
"omnifocus": {
"command": "omnifocus-mcp",
"args": []
}
}
}
That's it. The AI assistant now has full OmniFocus access.
What It Can Do
Tasks (23 tools)
Full lifecycle management for OmniFocus tasks:
- CRUD — create, get, update, delete individual tasks
- Batch operations — create, move, or delete multiple tasks in a single call
- Subtasks — create and list subtasks under any parent task
- Completion — mark complete, mark incomplete (supports repeating tasks)
- Search — full-text search across task names and notes with all filters applied
- Move and reparent — relocate tasks between projects, reparent tasks under other tasks, or move subtasks back to inbox/project without delete/recreate
- Duplicate — clone a task with all properties and optional subtasks
- Notifications — list, add, and remove notifications (absolute date or relative offset)
- Repetition — set or clear repetition rules with schedule type (regularly/after completion)
- Notes — append text to task notes without overwriting
- Safety model — destructive delete confirmations stay separate from non-destructive move/update workflows
- Aggregate counts — fast "how many" queries without listing individual tasks
Advanced Filtering
list_tasks and search_tasks support powerful filter combinations:
| Filter | Description |
| --- | --- |
| project | Scope to a single project by name |
| tag / tags | Filter by one tag or multiple tags |
| tagFilterMode | "any" (default) or "all" for multi-tag filtering |
| flagged | Flagged tasks only |
| status | "available", "remaining", "completed", "dropped", "all" |
| dueBefore / dueAfter | Due date range (ISO 8601) |
| deferBefore / deferAfter | Defer date range (ISO 8601) |
| completedBefore / completedAfter | Completion date range (ISO 8601) |
| plannedBefore / plannedAfter | Planned date range (ISO 8601) |
| maxEstimatedMinutes | Tasks with estimated duration up to N minutes |
Sorting
All list/search tools support sortBy and sortOrder:
- Sort by:
name,due,defer,added,modified,completed,estimated,planned - Sort order:
asc(default) ordesc
Projects (11 tools)
- CRUD — create, get, update, delete projects
- Lifecycle — complete, uncomplete, set status (active/on-hold/dropped)
- Organization — move between folders, search by name
- Filtering — by folder, status, completion date range, stalled-only flag
- Sorting — by name, due date, or other fields
- Aggregate counts — project counts by status, optionally scoped to a folder
Project Lifecycle Semantics
- Use
complete_projectwhen work is finished/closed (done/completed). - Use
set_project_statusfor organizational state only:active= currenton_hold= paused (UI wording is often "on hold"/"on-hold")dropped= intentionally abandoned/cancelled, not completed
- Use
uncomplete_projectto reopen a completed project back to active. - In user-facing summaries, present business meaning first (project name, folder, and status transition), and include opaque IDs only as secondary references.
Tags (5 tools)
- CRUD — create, update (name and status), delete
- List — with status filter (active/on-hold/dropped/all), sorting, and limits
- Search — fuzzy name matching
Folders (5 tools)
- CRUD — create, get (with child projects and subfolders), update, delete
- Hierarchy — create nested folders with parent parameter
- List — all folders with limits
Forecast (1 tool)
- Structured view with sections: overdue, due today, flagged, deferred, and due this week
Perspectives (1 tool)
- List all available OmniFocus perspectives
Resources (3)
Live snapshots available to MCP clients:
| Resource | Description | | --- | --- | | Inbox | Current inbox tasks | | Today | Today's forecast (overdue + due today + flagged) | | Active Projects | All active projects with task counts |
Prompts (4)
Ready-to-use review workflows:
| Prompt | Description | | --- | --- | | Daily Review | Due-soon, overdue, and flagged tasks for daily planning | | Weekly Review | Active projects and next-action coverage analysis | | Inbox Processing | One-by-one inbox clarification decisions | | Project Planning | Guided planning for a specific project |
Implementations
Three implementations with identical tool names, parameters, and response shapes:
| Implementation | Language | Install | Recommended For |
| --- | --- | --- | --- |
| Rust | Rust | Homebrew (recommended) or source | Production use — single binary, fast startup |
| Python | Python 3.11+ | uv from source | Local development, easy scripting |
| TypeScript | Node.js 20+ | npm from source | Node.js ecosystems |
Detailed setup guides: Rust · Python · TypeScript
How It Works
The server runs JXA (JavaScript for Automation) scripts through macOS osascript. Each script uses the OmniFocus evaluateJavascript bridge to execute Omni Automation JavaScript inside OmniFocus itself, where full APIs like flattenedTasks, Task.Status, and new Task() are available. Data is serialized as JSON and returned through the MCP protocol with consistent schemas across all three implementations.
MCP Client Config Examples
Claude Desktop
{
"mcpServers": {
"omnifocus": {
"command": "omnifocus-mcp",
"args": []
}
}
}
Cursor
{
"mcpServers": {
"omnifocus": {
"command": "omnifocus-mcp",
"args": []
}
}
}
Python (source build)
{
"mcpServers": {
"omnifocus": {
"command": "uv",
"args": ["run", "omnifocus-mcp"]
}
}
}
TypeScript (source build)
{
"mcpServers": {
"omnifocus": {
"command": "node",
"args": ["dist/index.js"],
"cwd": "/absolute/path/to/OmnifocusMCP/typescript"
}
}
}
Keep only one OmniFocus MCP server enabled at a time to avoid duplicate tool surfaces.
Compatibility snippet:
{
"mcpServers": {
"omnifocus": {
"command": "python",
"args": ["-m", "omnifocus_mcp"]
}
}
}
{
"mcpServers": {
"omnifocus": {
"command": "node",
"args": ["dist/index.js"],
"cwd": "/absolute/path/to/OmnifocusMCP/typescript"
}
}
}
Switching Implementations
Switching Between Rust, Python, and TypeScript
- Use Rust when you want a single prebuilt
omnifocus-mcpbinary. - Use Python when you want
uvorpython -mexecution and fast local iteration. - Use TypeScript when you want
nodeexecution fromtypescript/dist/index.js. - Restart the MCP client so it reloads the server command after you switch implementations.
Prerequisites
- macOS (required — OmniFocus is macOS-only)
- OmniFocus installed and running
- Automation permission granted to the terminal/editor (System Settings → Privacy & Security → Automation)
For source builds only:
- Python 3.11+ and
uv(Python implementation) - Node.js 20+ and npm (TypeScript implementation)
- Rust toolchain via
rustup(Rust source build)
Contributing
Contributions are welcome through focused pull requests with clear scope and passing checks. See CONTRIBUTING.md for setup and validation steps.
License
MIT. See LICENSE for details.