MCP Apple Reminders
A macOS-only MCP server that exposes Apple Reminders to AI clients, enabling create, read, update, delete, search, and organization of reminders and lists through natural language.
README
MCP Apple Reminders
A macOS-only Model Context Protocol server that exposes Apple Reminders to Claude Code, Codex, and the Claude Desktop app. It can create, read, update, delete, search, and organize reminders and lists, drive a Claude-* workflow board, manage sidebar groups, set alarms and recurrence, run bulk operations, and work with subtasks, sections, and tags.
The server is built on the MCP Python SDK (FastMCP) and reads/writes Apple Reminders through a three-tier native stack (direct SQLite reads + a Swift EventKit helper + an Objective-C ReminderKit helper). See Architecture.
Features
- 58 MCP tools across 17 modules (full catalog in
docs/TOOLS.md). - Calendar/list management — create, delete, update, search lists.
- Reminder CRUD — create, update, complete/uncomplete, delete, fetch by ID.
- Rich queries — today, overdue, next-up, completed-in-range, text search, filters.
- Workflow board — move reminders across Pierce's
Claude-*lists. - Sidebar groups — create/list/delete folders and move lists into them.
- Alarms & recurrence — time alarms, location alarms, repeat rules.
- Bulk operations — complete, move, and purge-completed in one call.
- Subtasks, sections & tags — hierarchy and grouping via the ReminderKit helper.
- MCP Resources, Prompts, Elicitation, Sampling, and progress reporting (see below).
- Deeplinks — every Reminder and Calendar carries an
x-apple-reminderkit://deeplink.
Requirements
- OS: macOS (EventKit-backed; macOS owns iCloud sync). No cross-platform support.
- Python: 3.10+ (the repo venv runs 3.13).
- Xcode Command Line Tools:
swiftc+clang, required once to compile the native helpers. - Permission: Reminders access, granted per interpreter binary (see Permissions).
- MCP client: Claude Code, Codex, the Claude Desktop app, or any MCP-compatible client.
Key dependencies: MCP Python SDK 1.27.x (FastMCP), Pydantic v2, and PyObjC/EventKit.
Installation
Recommended: install.sh
From the repo root:
./install.sh
This creates ./venv, installs the package in editable mode with its dependencies,
and compiles the native helpers (make build-native).
Manual
python3 -m venv venv
source venv/bin/activate
pip install -e ".[dev]"
make build-native # compiles the Swift + Obj-C helpers into _native/bin/
Verify the install
./venv/bin/python3 -c "from mcp_apple_reminders import cli_main; print('ok')"
./venv/bin/python3 verify_setup.py # preflight: interpreter, deps, perms, client configs
verify_setup.py triggers the macOS Reminders permission prompt on first run. Approve it.
Configuration
In every client, the launch command must be this repo's venv interpreter
(the binary that was granted Reminders access — see Permissions),
not a bare python3. On many macOS systems /usr/bin/python3 is too old anyway.
Replace /absolute/path/to/mcp-apple-reminders with your local checkout in each snippet.
Claude Code
claude mcp add apple-reminders \
/absolute/path/to/mcp-apple-reminders/venv/bin/python3 -- -m mcp_apple_reminders
Codex
Codex reads MCP servers from ~/.codex/config.toml:
[mcp_servers.apple-reminders]
command = "/absolute/path/to/mcp-apple-reminders/venv/bin/python3"
args = ["-m", "mcp_apple_reminders"]
cwd = "/absolute/path/to/mcp-apple-reminders"
enabled = true
Claude Desktop app
Edit ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"apple-reminders": {
"command": "/absolute/path/to/mcp-apple-reminders/venv/bin/python3",
"args": ["-m", "mcp_apple_reminders"],
"env": {}
}
}
}
Restart Claude Desktop, then look for the tools under "apple-reminders".
First-run permission shim
If the Reminders permission prompt never appears, point the client's command at
./shim_mcp.sh (args []) once to surface the prompt, then switch back to the venv
interpreter above.
Other clients
The server speaks MCP over stdio. The general invocation is:
/absolute/path/to/mcp-apple-reminders/venv/bin/python3 -m mcp_apple_reminders
Permissions
Apple Reminders access is granted per binary, so the exact interpreter the client
launches (the conda/venv python3) must be the one that holds the grant.
- On first launch macOS shows a permission dialog — click OK.
- If you miss it, run
./venv/bin/python3 verify_setup.py(or./shim_mcp.sh) once and approve. - To inspect/toggle manually: System Settings → Privacy & Security → Reminders.
Without the grant the server cannot read or write reminders.
Tools, Resources & Prompts
The 58 tools are grouped into 17 modules under src/mcp_apple_reminders/tools/.
The exhaustive, always-current catalog (parameters + return shapes) lives in
docs/TOOLS.md. Summary by category:
| Category | Count | What it does |
|---|---|---|
| calendars | 8 | create/delete/update lists, get by name/ID, default, list, search |
| reminders | 4 | create, update, delete, get by ID |
| completion | 2 | complete / uncomplete shortcuts |
| queries | 7 | get/filter, today, overdue, next-up, completed-in-range, search, recently-deleted |
| workflow | 6 | move across Claude-* board lists; list the board |
| groups | 4 | create/list/delete sidebar folders; move a list into a group |
| alarms | 3 | time alarm, location alarm, recurrence rule |
| bulk | 3 | bulk complete, bulk move, bulk delete-completed |
| sections | 3 | get subtasks, set parent, assign section |
| smartlists | 3 | create/update/delete custom smart lists |
| appearance | 3 | list/smart-list appearance + pinning |
| templates | 3 | create/apply/delete list templates |
| flags | 3 | set urgent, set early reminder, add section + assign |
| attachments | 3 | add URL / metadata / file attachments |
| grocery | 1 | categorize_grocery_items |
| agents | 1 | bootstrap_agent_list — agent-visibility plane |
| sampling | 1 | triage_brain_dump — server-initiated LLM sampling |
The workflow board is Pierce's Claude-* list family: Claude-Brain-Dump,
Claude-On-Deck, Claude-Active, Claude-Waiting, and Claude-Done. Groups are
Reminders.app sidebar folders.
Beyond tools, the server also exposes:
- MCP Resources (read-only views):
reminders://default,reminders://today,reminders://overdue,reminders://list/{calendar_id},reminders://recently-deleted,reminders://tags,reminders://appearance, andagents://current/{project_name}(the agent-visibility plane). - MCP Prompts:
daily_review,weekly_retro,brain_dump_triage,agent_visibility_sync,organize_into_sections. - Elicitation guards on destructive operations (e.g. deletes/purges).
- Server-initiated Sampling via
triage_brain_dump. - Progress reporting on long-running operations.
Usage examples
Once configured, ask your client in natural language — it picks the tools:
- "Show me everything due today." →
get_today_reminders - "Remind me to call the dentist tomorrow at 2pm." →
create_reminder - "What am I behind on?" →
get_overdue_reminders - "Move the 'update docs' task to Active." →
move_reminder_active - "Triage my brain dump." →
triage_brain_dump(with elicitation/sampling)
Architecture
server.py is the FastMCP instance; importing the tools, resources, and prompts
packages registers everything against it. lifespan.py builds the AppContext
(SQLite store path + helper binary paths); models.py holds frozen Pydantic v2 models,
each Reminder/Calendar carrying a deeplink (e.g. x-apple-reminderkit://REMCDReminder/{uuid}).
All native access lives under src/mcp_apple_reminders/_native/ in three tiers:
- SQLite reads (
sqlite.py) — reads the Reminders.app CoreData SQLite store directly. Fast path for reads, including sections, subtasks, tags, and groups. - EventKit writes (
eventkit.py) — Python wrapper around a compiled Swift EventKit helper subprocess for public-API writes. Binary_native/bin/rem_eventkit, compiled from_native/src/rem_eventkit.swift. - ReminderKit writes (
reminderkit.pytransport +reminderkit_actions.pytyped wrappers) — Python wrapper around a compiled Objective-C ReminderKit (a private framework) helper subprocess for subtasks, tags, sections, flagged state, and groups. Binary_native/bin/rem_reminderkit, compiled from_native/src/rem_reminderkit.m.
Both native helpers are borrowed from viticci/remctl
(MIT); see src/mcp_apple_reminders/_native/THIRD_PARTY_NOTICES.md.
Client (Claude Code / Codex / Claude Desktop)
| MCP over stdio (JSON-RPC)
server.py (FastMCP: tools + resources + prompts)
| lifespan AppContext (sqlite path, helper paths)
_native/ ── sqlite.py ........... direct CoreData SQLite reads
├ eventkit.py ........ Swift helper subprocess (public-API writes)
└ reminderkit.py ..... Obj-C helper subprocess (private-framework writes)
|
Apple Reminders store
No stdout, ever. stdio is the JSON-RPC transport; any stray print corrupts the
protocol. All logs go to stderr only.
Development
# Lint, format-check, line-limit gate
ruff check src/ test_*.py test_support/
black --check src/ test_*.py test_support/
make check-architecture # hard cap 400 lines/file
# Tests (root-level, explicit paths — root is not on testpaths)
./venv/bin/python -m pytest test_mcp_tools.py test_workflow_tools.py test_e2e.py
# Rebuild native helpers after editing the Swift/Obj-C sources
make build-native # or: make clean-native && make build-native
Run the server directly for debugging (logs to stderr):
./venv/bin/python3 -m mcp_apple_reminders
The installed console script mcp-apple-reminders maps to mcp_apple_reminders:cli_main.
Troubleshooting
- "No access to reminders" — the grant is per binary. Confirm the client launches
the venv interpreter and that that binary is enabled under System Settings →
Privacy & Security → Reminders. Re-run
verify_setup.pyor./shim_mcp.shto re-prompt. - Tools missing in the client — verify the config path and JSON/TOML syntax, confirm the interpreter path is absolute, and fully restart the client.
ModuleNotFoundError— reinstall:pip install -e ".[dev]".- Native helper missing / write failures — run
make build-native(needs Xcode CLI tools). Check_native/bin/rem_eventkitand_native/bin/rem_reminderkitexist. - DateTime errors — use ISO 8601, e.g.
2026-01-15T14:30:00. - Reminder not found — the ID may be stale; re-fetch via
search_reminders/get_reminders.
Client logs (Claude Desktop): tail -f ~/Library/Logs/Claude/mcp*.log.
Further docs
docs/TOOLS.md— full tool catalog.docs/MAP.md— module map.docs/SQLITE_SCHEMA.md— Reminders SQLite schema notes.docs/SECURITY-REVIEW.md— security review.AGENTS.md— agent collaboration guide.CHANGELOG.md— release history.
License
MIT. See LICENSE. Native helpers are MIT-licensed third-party code — see
src/mcp_apple_reminders/_native/THIRD_PARTY_NOTICES.md.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.