
Apple Notes Snapshot
io.github.xiaojiou176-open由 GitHub 帳號 xiaojiou176-open 發布v0.1.12更新於 Oct 24, 2026
Local-first Apple Notes backup control room for macOS with a read-only MCP surface.
概覽
一個本機、唯讀優先的 MCP 介面,向助理揭露 macOS 上 Apple Notes 備份控制室的診斷與狀態資訊。
- 可以做什麼
- Apple Notes Snapshot 是圍繞 notes-exporter 引擎打造的 macOS 備份控制室,其 MCP 介面是面向代理的 stdio 優先、唯讀優先通道。它揭露本機備份診斷與資源,例如快照健康狀態、上次成功時間、launchd 狀態、失敗原因、doctor 警告以及近期執行摘要。它不會寫回 Apple Notes,也不是託管服務。
- 適用情境
- 當你已在 macOS 上依賴 Apple Notes,並希望助理讀取與人工操作者相同的本機備份事實時使用。它適合能啟動 stdio MCP 伺服器並需要備份健康狀態脈絡的本機編碼代理工作流程。不適用於雲端同步、團隊協作或雙向寫回。
- 執行需求
- 僅限 macOS 桌面,以本機程序執行。MCP 介面透過 notesctl 指令啟動,底層工作流程需要 Apple Notes 與 AppleScript 權限,並需先完成一次成功的快照執行。清單中未宣告驗證、環境變數或標頭。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 Apple Notes Snapshot,將其新增到工作區。
- 為需要使用其工具的對話啟用該服務。
Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。
其他 MCP 客戶端
參照 儲存庫 中的啟動說明。
README
Apple Notes Snapshot
Apple Notes Snapshot is the backup control room for Apple Notes on
macOS. Keep the upstream notes-exporter engine, then add path-aware exports,
launchd scheduling, visible health checks, and a calmer recovery path when
the local backup loop drifts.
🤝 Sister repo: looking to recover deleted notes after the fact, not back them up beforehand? See notes-recover — the forensic-grade reader for Apple Notes'
NoteStore.sqlite(recover deleted notes, browse the raw store, ship audit-ready bundles).notebackupcovers the before;notes-recovercovers the after. Together they cover the full Apple Notes lifecycle on macOS.
After the control room already makes sense, the repo also ships extra builder and diagnostics surfaces so humans and coding agents can inspect the same backup state without turning the workflow into a hosted service.
Start the 3-step quickstart | First-run troubleshooting | Open the proof page | Compare with upstream | Browse release history | Get support or routing help
[Trusted CI] [Latest Release] [macOS only]
[Apple Notes Snapshot Web console showing run health, scheduler status, and quick actions]
- Keep the path obvious Review or override the export destination before the first snapshot writes.
- Turn one-off exports into a local loop
Use
launchdto schedule repeatable snapshots instead of babysitting manual exports. - See health before you guess Check freshness, failure reasons, logs, and recovery clues when the loop drifts.
- Read the next safe move first The Web console now surfaces an operator focus deck above the raw transcript so you can see the next move, the reason, and the reading order before diving into action output.
Category: Apple Notes backup control room. AI/agent hook: AI-assisted diagnostics plus optional coding-agent access once the operator workflow is already clear. Result: a calmer, more reviewable backup workflow on your own Mac.
Best fit: people who already rely on Apple Notes and want a calmer, reviewable backup routine on their own Mac.
Not the goal: cloud sync, team collaboration, or two-way write-back into Apple Notes.
At A Glance
If you only want the shortest truthful filter before reading deeper, use this table first:
What you can prove in one local pass
- The destination is obvious: you can review or override the snapshot path before anything writes.
- The loop is real: one manual run plus
launchdturns a one-off export into a repeatable local rhythm. - The control room is inspectable:
status,verify,doctor, logs, and proof pages all stay on the same local facts. - Integration surfaces are optional: AI Diagnose, the Local Web API, and MCP all stay behind the operator story instead of replacing it.
If you want the shortest public evidence trail after that first pass, open the proof page. It collects the repo-owned gates, the GitHub-controlled release and Pages evidence, and the current access boundary in one place.
Start with Run -> Install -> Verify
Treat the first healthy loop like a three-stop checklist, not like a dashboard to decode all at once.
- Run
Use
./notesctl run --no-statusonce so macOS can surface permissions and the local ledger can record a first successful baseline. - Install
Use
./notesctl install --minutes 30 --loadafter that first run so the workflow becomes a repeatablelaunchdloop instead of a manual chore. - Verify
Use
./notesctl verifyfirst, then./notesctl doctoronly if warnings or empty state remain.
After that verified loop exists, the Web console, proof page, AI Diagnose, Local Web API, and MCP surfaces become much easier to read because they are all describing a baseline you already proved.
Integration and maintainer routes after the operator path
The public product front door is still the local backup control room. The lanes
below are for builders after the first healthy loop exists; they do not replace
Run -> Install -> Verify.
- Public product front door: backup control room
- the main story is still the local path review, the first snapshot run, the
launchdloop, and the health check surface on your own Mac
- the main story is still the local path review, the first snapshot run, the
- Second ring builder protocol lane: pure MCP
- the repo's main machine-facing surface is the local stdio MCP flow behind
./notesctl mcpand the rootserver.json
- the repo's main machine-facing surface is the local stdio MCP flow behind
- Second ring builder packet lane: pure skills
- the public skill packet at
examples/public-skills/notes-snapshot-control-room/teaches hosts how to attach to that same local control room
- the public skill packet at
- Companion later lanes: plugin shell and
.mcpbpackagingplugins/apple-notes-snapshot-control-room/andpackaging/mcpb/are host-shaped packaging surfaces around the same local workflow- they matter, but they are not the first story a new reviewer should read
- Current non-claims
- no hosted Notes service
- no cloud sync product
- no official skill or plugin marketplace listing without fresh host-side read-back
- no cross-machine attach guarantee
Use the integration links below only after the operator loop already makes sense.
Secondary integration reads after the first healthy loop: AI Diagnose | Local Web API | MCP Provider | Distribution and listing boundaries | For Codex / Claude Code integrations
Quickstart
Start here if you want the shortest honest path from manual export to a repeatable local snapshot loop. The first-successful-run promise is 3 steps and about 3 minutes, not a zero-click install.
- Review
config/notes_snapshot.envand keep or change the default export path. - Run one snapshot so macOS can show Apple Notes / AppleScript permission prompts.
- Install the scheduler, then verify health.
Fail fast:
- If
./notesctl verifysaysFAIL: no last_success record; run ./notesctl run, you have not completed the first manual export yet. - If macOS permissions block the first run, use
./notesctl permissionsand the public troubleshooting guide. - The full, authoritative guide lives at docs/quickstart.
- The public proof page shows the repo-side gates, live-surface checks, and trust boundary in one place.
- Maintainer-grade verification lives later in Proof and verification; it is not required for the first successful snapshot.
Choose the right lane
Keep the operator lane and the maintainer lane separate.
- Operator lane
./notesctl run --no-status./notesctl install --minutes 30 --load./notesctl verify./notesctl doctor./notesctl status --full- optional after the local state exists:
./notesctl ai-diagnose,./notesctl web, and./notesctl mcp
- Maintainer lane
./notesctl clean-cache --dry-run./notesctl clean-cache./notesctl rebuild-dev-env./notesctl update-vendor./notesctl setup/./notesctl self-heal
If your goal is a healthy backup loop, stay in the operator lane first. The maintainer lane is for repo upkeep and contributor verification after the local workflow already makes sense.
Why this exists
The upstream notes-exporter
project is great when you want to export Apple Notes right now. This repository
exists for the moment when "run it once" turns into "keep it healthy every
day."
Apple Notes Snapshot wraps the upstream exporter with:
- one supported human entrypoint:
notesctl - scheduled execution via
launchd - lock protection to avoid overlapping runs
- structured state and metrics files for health checks
- log rotation and log-health summaries
- an optional local Web console for status and safe actions
What you get over upstream notes-exporter
Think of upstream as the engine and this repository as the control room around it. The wrapper tells you where snapshots will land, and makes the schedule and health surface easier to inspect when something breaks.
[Comparison card showing the difference between upstream notes-exporter and Apple Notes Snapshot]
Use upstream directly if you only need a one-time export. Use this repository when you want a repeatable local backup loop with visible scheduler state, clear snapshot paths, and easier recovery when something fails.
What the Web console shows
The local Web console is optional, and it makes the most sense after you have
already completed the first run -> install -> verify pass. Use it to watch
the loop you just proved, not as a replacement for that first proof.
It surfaces:
- snapshot health, last success, launchd state, and failure reason
- an operator focus deck that says what to do next and which panel to open first
- doctor warnings and dependency readiness
- recent metrics and trigger sources
- log-health summaries
- access policy and quick actions
[Architecture diagram for Apple Notes Snapshot]
Optional AI and agent-facing surfaces
These are additive surfaces around the same local control room. They do not
replace notesctl or the deterministic runtime checks.
- AI Diagnose
Use
./notesctl ai-diagnosewhen you want an advisory explanation of the current local state. It readsstatus,doctor,log-health, and recent-run summaries, routes model calls through a local Switchyard runtime when AI is enabled, and still works as a deterministic fallback when no AI provider is configured. Read the public AI Diagnose guide. - MCP Provider
Use
./notesctl mcpwhen you want a stdio-first, read-only-first MCP surface for agents. It exposes local backup diagnostics and resources without turning the Web console into a fake MCP API. Read the public MCP guide. - Local Web API
Use
./notesctl webwhen you want the token-gated local browser control room plus a small JSON API backed by the same repo-owned command surface. It is a local operator API, not a public OpenAPI or hosted integration surface. Read the public Local Web API guide.
How AI and agents fit
Use this mental model if you care about Codex, Claude Code, MCP, or other host-local integration ecosystems.
Natural fit:
-
Codex- and Claude Code-style local workflows when the host can launch stdio MCP servers
-
MCP-aware coding agents that need the same local backup facts as a human operator
-
AI Diagnose = operator next-step assistant
- It explains local status, doctor, log-health, and recent-run evidence.
- It is not a generic chat overlay and it does not become the system truth.
-
MCP Provider = read-only agent substrate
- It exposes the same local backup state to MCP-aware hosts.
- It does not turn the project into a hosted agent platform or write-capable remote control plane.
-
Local Web API = token-gated browser/API surface
- It serves the local Web console plus JSON endpoints like
status,doctor,recent-runs, andaccess. - It is meant for browser or local HTTP workflows on your own machine, not as a public OpenAPI promise.
- It serves the local Web console plus JSON endpoints like
-
notesctl+state.json+ aggregate summaries + token-gated Web API = current substrate- This repository does not ship a public OpenAPI, generated client, or SDK today.
- The truthful integration entry points are the CLI contract, the web/API surface, and the read-only MCP surface.
Integration-facing docs have their own shelves now, so this README does not need to carry every host-specific setup detail:
- Open the public For Agents guide for the truthful builder overview and proof legend.
- Open the Codex starter pack, Claude Code starter pack, or OpenClaw starter pack when you need host-shaped install guidance.
- Open the Builder integration pack for the capability matrix and copyable examples, or the Public skills pack for the curated public-safe instruction subset.
- Open the Local Web API guide if your workflow is browser- or local-HTTP-shaped instead of stdio MCP.
Typical workflows
[Three-step run flow showing rebuild, first manual export, and scheduler verification]
1. Review the export path, then prove the first run
This is the primary getting-started path. Everything else in the docs site is a supporting surface around that flow.
2. Check whether backups are still healthy
3. Open the local console when you want a control room
4. Refresh the vendored upstream when you maintain the wrapper
Who this is for
- People who already rely on Apple Notes and want local, repeatable snapshots
- macOS users who prefer
launchdover cloud schedulers - Anyone who wants backup health, logs, and operational visibility around Apple Notes exports
- Maintainers who want a reviewable wrapper instead of ad hoc shell snippets
Who this is not for
- Teams looking for a shared cloud notes product
- Anyone expecting two-way sync back into Apple Notes
- Cross-platform note automation workflows
- Hosted SaaS or desktop app buyers
Proof and verification
This repository ships with repo-owned verification commands instead of asking you to trust screenshots alone. These are maintainer / contributor gates, not required for the first successful snapshot.
If you want the shorter public-facing evidence page first, open the proof page. It keeps the repo-side gates, GitHub-controlled delivery facts, and the current access boundary in one place. The ladder below remains the maintainer-grade verification contract.
Default local maintainer lane:
Maintainer verification expects a local Python 3.11+ toolchain. ./notesctl rebuild-dev-env
recreates .runtime-cache/dev/venv from scratch so the documented verification
commands stay aligned with the current checkout path and interpreter.
GitHub Actions for this repo run on GitHub-hosted runners; the local ladder
below is maintainer verification, not a self-hosted runner requirement.
Five-layer CI contract:
GitHub-only governance gates:
Dependency Reviewruns on pull requests because it needs GitHub's base/head dependency diff.CodeQL,Secret Scan,GitHub Alert Gate,Actionlint,Zizmor, andTrivystay hosted-first; local reruns are optional maintainer repro steps, not part of the default hook path.- Latest release tags should point at the current canonical closeout commit; do not treat an older tag as proof of current repo-side closure.
Runtime and cache contract
Think of the runtime layout as two rooms with one job each:
- repo-local rebuildables live under
.runtime-cache/ - repo-owned machine-level residue lives under the current machine cache root
managed by
notesctl
Current repo-local contract:
.runtime-cache/dev/venv-> repo-owned maintainer virtual environment.runtime-cache/cache/apple-notes-snapshot-> repo-local runtime cache/state support.runtime-cache/temp-> scratch.runtime-cache/logs-> repo-local logs.runtime-cache/pytest-> pytest cache.runtime-cache/coverage-> coverage data.runtime-cache/pycache-> Python bytecode cache.runtime-cache/browser-proof-> generated proof screenshots that can be rebuilt from the current docs surface.runtime-cache/phase1-> historical hard-cut rollback artifacts, not current runtime truth.runtime-cache/phase1-history-rebuild-> historical rebuild rollback artifacts, not current runtime truth.runtime-cache/mcp-registry-lane/out-> release-ready MCP registry artifacts that can be rebuilt on demand
Current external repo-owned contract:
- launchd records for current and stale labels
- runtime residue for launched repo-owned services
- repo-scoped runtime copies used by the launchd wrapper
- the persistent isolated browser root for this repo
- disposable browser temp residue
- the vendor-runtime current pointer/cache
Older machine-level Application Support and cache roots are migration inputs only. New repo-owned runtime/cache writes should stay inside the current repo-managed machine cache root.
Local cleanup and runtime audit
This is a maintainer-only cleanup lane. It is useful when you are rebuilding verification tooling or reclaiming repo-owned runtime residue, not when you are trying to complete the first successful snapshot.
./notesctl clean-cache --dry-run- previews repo-local rebuildable/disposable cleanup under
.runtime-cache/
- previews repo-local rebuildable/disposable cleanup under
./notesctl clean-cache- removes repo-local rebuildables, generated proof captures, historical rollback bundles, and disposable-generated residue
./notesctl runtime-audit- reports repo-local support surfaces plus external repo-owned cache/runtime roots
./notesctl clean-runtime --dry-run- previews cleanup for stale non-current residue under the repo-managed machine cache root
./notesctl clean-runtime- removes stale external repo-owned residue while protecting current launchd labels
./notesctl browser-bootstrap- copies the current
apple-notes-snapshotsource profile out of the default Chrome root into the repo-owned isolated root
- copies the current
./notesctl browser-open- launches or attaches to the single repo-owned Chrome instance for this repo
./notesctl rebuild-dev-env- restores the canonical path-aware maintainer environment after cleanup
The repo may still clean legacy .pytest_cache, .coverage, or scattered
__pycache__ directories if they are already present, but those are migration
backstops. The current contract routes repo-owned disposable artifacts into
.runtime-cache/*, including historical rollback folders and proof captures
that are safe to regenerate when you no longer need them.
Automatic janitor hooks run on repo-owned entrypoints that create or reuse
machine-level residue, including run, web, install, ensure,
rebuild-dev-env, and runtime-audit. The default policy is intentionally
strict:
- external repo-owned budget:
2 GB - general external TTL:
72 hours - browser clone TTL:
24 hours - current/protected launchd labels are never deleted by TTL alone
browser/chrome-user-data/is permanent state and excluded from TTL/cap cleanup
This repository does not have a repo-owned Docker cleanup lane today. The automatic janitor does not touch Docker, system temp roots, or shared tool caches from Cursor, Codex, Claude, Serena, Homebrew, pip, nodeenv, uv, or other machine-wide tooling.
Browser automation contract
Browser automation in this repository now uses an isolated Chrome root + single repo-owned instance + CDP attach contract.
NOTES_SNAPSHOT_BROWSER_PROVIDER=chromeNOTES_SNAPSHOT_BROWSER_ROOT=<repo-owned-browser-root>NOTES_SNAPSHOT_CHROME_USER_DATA_DIR=<repo-owned-browser-root>/chrome-user-dataNOTES_SNAPSHOT_CHROME_PROFILE_NAME=apple-notes-snapshotNOTES_SNAPSHOT_CHROME_PROFILE_DIR=Profile 1NOTES_SNAPSHOT_CHROME_CDP_HOST=127.0.0.1NOTES_SNAPSHOT_CHROME_CDP_PORT=9337
The old default Chrome user-data root is now only a one-time read source for
./notesctl browser-bootstrap. It is no longer the long-term runtime root for
this repo.
Treat browser-bootstrap as a one-time migration, not as a routine sync step.
If you later add or refresh logins inside the isolated root, do not rerun
browser-bootstrap unless you intentionally want to replace the isolated root
from the default Chrome root again.
Use these commands in order:
./notesctl browser-bootstrap- one-time copy from the default Chrome root into the isolated repo-owned root
./notesctl browser-open- launch the single repo-owned Chrome instance if it is not running yet
- otherwise return attach info instead of second-launching
./notesctl browser-contract --json- print the canonical attach-first contract, including the CDP endpoint for Playwright/CDP clients
If 127.0.0.1:9337 is already occupied on your machine, the repo should fail
fast instead of silently attaching to the wrong thing. In that case, use a
deliberate env override such as NOTES_SNAPSHOT_CHROME_CDP_PORT=9347 before
starting or attaching to the repo-owned instance.
This repo does not silently fall back to bundled Chromium, and it does not keep launching new browser instances against the same user-data dir. Human manual use and automation are expected to attach to the same repo-owned Chrome instance.
CI contract
The repository keeps five verification layers, and they are not interchangeable:
This open-source repository does not rely on a local self-hosted runner lane. Local verification exists so maintainers can reproduce the repo-owned gates before or after a pull request, not because CI is expected to run on the developer's Mac.
High-value local checks:
The repo-owned quick gate now covers docs/root hygiene, vendor tree hygiene,
unit tests, and wrapper-level JSON/help smoke checks. GitHub alert state moved
fully into the hosted lane so the default local pre-push path stays lighter and
more deterministic. The 90%+ coverage bar still applies to repo-owned Python
surfaces under scripts/ops; the shell wrapper surface is guarded by smoke
checks rather than pretending it shares that coverage metric.
Verification ladder and test pyramid
Think of the verification contract like a small testing pyramid instead of one giant "just run everything" blob.
- Unit tests
- fastest checks for repo-owned Python logic and deterministic report shaping
- examples:
tests/unit/test_ai_diagnose_unit.py,tests/unit/test_mcp_server_unit.py,tests/unit/test_web_server_unit.py
- Wrapper smoke
- repo-owned shell-contract checks for
notesctlhelp, JSON surfaces, and wrapper entrypoints - example:
bash scripts/checks/run_wrapper_smoke.sh
- repo-owned shell-contract checks for
- E2E
- slower integration checks for launchd, relocation, Web server behavior, and real command orchestration
- example:
./.runtime-cache/dev/venv/bin/pytest tests/e2e --no-cov
- Manual-local validation
- the real operator path on a real Mac:
run -> verify -> status -> doctor -> web
- the real operator path on a real Mac:
Read the pyramid as:
- unit tests catch logic regressions fastest
- wrapper smoke protects the shell contract
- E2E checks prove the repo-owned pieces still cooperate
- manual-local validation proves the actual machine is healthy
Security and privacy
This project is intentionally user-controlled on your own machine:
- your Apple Notes content stays in your own Apple Notes account and export destination
- the optional Web console is local by default and token-aware when enabled
- the repository documents private vulnerability reporting in
SECURITY.md - secrets, runtime caches, generated launchd files, and logs are not part of the tracked source surface
Read SECURITY.md before reporting sensitive issues.
FAQ
Is this a replacement for the upstream exporter?
No. It is a wrapper that makes the upstream exporter safer and easier to operate as an ongoing local workflow.
Does it sync changes back into Apple Notes?
No. This repository is for snapshotting and exporting, not two-way sync.
Do I need the Web console?
No. notesctl is the supported human entrypoint. The Web console is optional.
How do I open the interactive menu?
Run ./notesctl with no arguments. The interactive menu lives on the canonical
command surface now.
Do I need GitHub Pages to use the tool?
No. Pages is only for public documentation and search visibility. The actual workflow stays local on your Mac.
What if I move the checkout to another path?
Reinstall the scheduler so the path-aware launchd surface points at the current checkout. If you also use the repo-owned verification environment or the optional Web console, rebuild that environment too:
Community
Use the GitHub Discussions surface that matches your intent:
- Questions and setup blockers: Q&A discussion
- Feature ideas and roadmap input: Ideas discussion
- Real-world usage examples: Show and tell discussion
- Release announcements: Announcements discussion
If you need a support-routing summary first, read SUPPORT.md or the public support page.
Contributing
Read CONTRIBUTING.md before changing wrapper logic, docs, or public-facing copy. The project favors small, reviewable changes with explicit verification notes.
License
This repository is released under the MIT License.
來源:README.md,提交 e56150b
工具
0版本歷史
1- v0.1.12最新Sep 16, 2026
評分
已滿足適用檢查項的 50%
該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。
- 近 30 天內可連通25 分
請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。
- 已宣告認證方式10 分
請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。
- 已列出工具15 分
請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。
- 描述必需15 分
- 圖示10 分
- 文件15 分
- 授權條款10 分