AI Decision System
Replayable, offline-verifiable records for AI-assisted go/no-go decisions: experiment promotions, typed model answers and human-approved actions.
OmniFocus 연동 MCP 서버, Claude·다른 MCP 클라이언트가 OmniFocus 프로젝트와 상호작용
MCP server for OmniFocus integration - enables Claude and other MCP clients to interact with OmniFocus projects
개발 도구Python · ShellGitHub @s-morgan-jeffries직접 설치운영자 미확인
써 보셨어요? 한 번만 눌러 주세요. 로그인은 필요 없어요.
README 전체를 불러오는 중입니다. 아래는 저장된 발췌입니다.
A comprehensive, fast, reliable, and agent-friendly MCP server for OmniFocus on macOS.
21 tools covering projects, tasks, folders, tags, perspectives, and focus. Full CRUD with unified batch operations (Pydantic model inputs), 20+ filter types on task and project queries, and date management including recurrence (RRULE read/write).
Sub-second reads on filtered queries, even with hundreds of tasks in the database.README 전문 보기 ↗A comprehensive, fast, reliable, and agent-friendly MCP server for OmniFocus on macOS.
21 tools covering projects, tasks, folders, tags, perspectives, and focus. Full CRUD with unified batch operations (Pydantic model inputs), 20+ filter types on task and project queries, and date management including recurrence (RRULE read/write).
Sub-second reads on filtered queries, even with hundreds of tasks in the database.
| Operation | Time | Items | Database |
|---|---|---|---|
| Get flagged tasks | 0.66s | 14 | 202 tasks |
| Get overdue tasks | 0.69s | 8 | 202 tasks |
| Get next tasks | 0.66s | 27 | 202 tasks |
| Get inbox tasks | 0.64s | 10 | 202 tasks |
| Search tasks by keyword | 1.07s | 37 | 202 tasks |
| Get all tasks (unfiltered) | 2.20s | 202 | 202 tasks |
| Get all projects | 0.57s | 35 | 35 projects |
| Create or update a task | 0.9s | — | — |
Full profiling data: PERFORMANCE_PROFILING.md
93% code coverage from 1027 unit tests. 192 integration and E2E tests run against real OmniFocus, plus 5 protocol-level smoke tests verifying the full MCP JSON-RPC transport. Tests cover task, project, and tag lifecycles, filtering, hierarchy, dates, recurrence, and review workflows.
73-scenario blind eval suite with frontier models scoring 100% and popular open-weight models scoring 90-94% (full results). Agents that have never seen OmniFocus can correctly use every tool from descriptions alone. Scenarios cover tool selection, parameter usage, multi-step workflows, date semantics, recurrence, tag behavior, task movement, text search, and safety-critical operations (drop vs delete, destructive action guardrails). Server instructions teach GTD concepts (task states, project types, sequential dependencies, review cycles) so agents make informed decisions, not just API calls.
git clone https://github.com/s-morgan-jeffries/omnifocus-mcp.git
cd omnifocus-mcp
git checkout v0.13.2 # Latest stable release
# Install uv (if not already installed)
curl -LsSf https://astral.sh/uv/install.sh | sh
# Install dependencies
uv sync --dev
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"omnifocus": {
"command": "uv",
"args": [
"--directory",
"/absolute/path/to/omnifocus-mcp",
"run",
"omnifocus-mcp"
]
}
}
}
python -m omnifocus_mcp.server_fastmcp
Communicates via stdin/stdout using the MCP protocol.
macOS will prompt for:
Bug reports and feature requests are welcome via GitHub Issues.
MIT