simplenote-mcp
A remote MCP connector that gives Claude.ai projects a persistent, cross-conversation file store, backed by Simplenote and deployed free on Cloudflare Workers.
README
simplenote-mcp
A remote MCP connector that gives Claude.ai projects a persistent, cross-conversation file store, backed by Simplenote and deployed free on Cloudflare Workers.
The problem
A Claude.ai project conversation has no writable storage that survives across conversations — confirmed by probing the project sandbox from inside it:
| Surface | Agent can write? | Crosses conversations? |
|---|---|---|
/mnt/project |
no (only the user adds files) | yes |
/mnt/user-data/outputs |
yes | no — keyed per conversation |
VM (/home, /tmp) |
yes | no (ephemeral) |
/mnt/user-data/outputs is durable within a conversation, but its storage
namespace is the conversation id — a file written in one conversation is
invisible in the next, even inside the same project. The only cross-conversation
surface, /mnt/project, is read-only to the agent. So there is no
writable and cross-conversation surface — and because the sandbox's network
egress is allow-listed, a script running inside it can't reach Simplenote
directly either.
This connector supplies the missing surface from outside the sandbox: each
conversation pulls its files at the start and pushes changes back before ending.
The pull→work→push protocol the in-conversation agent follows lives in
BOOTSTRAP.md — add that file to your Claude project.
Why Cloudflare Workers
A connector must answer instantly when Claude calls a tool, which rules out
free hosts that sleep. Workers never sleep (~5 ms cold start), the free tier
needs no credit card, HTTPS is built in, and it's Anthropic's reference path
for remote MCP with OAuth (workers-oauth-provider + McpAgent). SQLite
Durable Objects — which McpAgent uses for session state — are now on the free
plan, so the whole thing runs for $0. (The sandbox's allow-listed egress can't
reach Simplenote, so the connector has to be remote anyway.)
The Simplenote/Simperium client is a faithful TypeScript port of the public
simplenote.py library (token auth,
the first-line-heading→filename convention, version-aware writes, markdown
system tag), so notes round-trip with the Simplenote apps.
Model
One Simplenote note per file, scoped by a project tag
(claude-project-<id>) that isolates a project's notes from the rest of the
account. A file's path is the slug of its first-line heading
(# Design Notes → Design-Notes.md).
| Tool | Purpose |
|---|---|
list_files |
discover persisted files |
read_file(path) |
pull one file |
write_file(path, content) |
persist/overwrite a file |
delete_file(path) |
trash a file |
Single-user by design: one Simplenote account (token stored as a Worker secret); OAuth just gates access to you with a shared password.
Deploy
Prereqs: a free Cloudflare account, pnpm + Node, and a
Simplenote API token. Mint one from your Simplenote email/password with the
public simplenote library:
pip install simplenote
python -c "from simplenote import Simplenote; print(Simplenote('EMAIL', 'PASSWORD').get_token())"
pnpm install
# 1. Log in to Cloudflare (opens a browser):
pnpm exec wrangler login
# 2. Create the KV namespace the OAuth provider uses, then paste the printed id
# into wrangler.jsonc (kv_namespaces[0].id).
pnpm exec wrangler kv namespace create OAUTH_KV
# 3. (optional) set your project tag in wrangler.jsonc → vars.SIMPLENOTE_PROJECT_TAG
# 4. Set secrets (not in any file):
pnpm exec wrangler secret put SIMPLENOTE_TOKEN # your Simperium token
pnpm exec wrangler secret put ACCESS_PASSWORD # a password you choose
# 5. Ship it:
pnpm run deploy
You'll get https://simplenote-mcp.<your-subdomain>.workers.dev.
Connect in Claude.ai (Max/Team/Enterprise): Settings → Connectors → Add
custom connector → URL https://simplenote-mcp.<subdomain>.workers.dev/mcp.
Claude runs the OAuth flow; enter your ACCESS_PASSWORD on the login screen.
Finally, upload BOOTSTRAP.md into the project's files
(open your project → Add files) so every conversation picks up the protocol.
Then, in a chat, ask Claude to load my project files from Simplenote to
rehydrate at the start, and save these back to Simplenote to persist before
you finish.
Local development
cp .dev.vars.example .dev.vars # fill in SIMPLENOTE_TOKEN + ACCESS_PASSWORD
pnpm dev # wrangler dev (local Worker)
pnpm dlx @modelcontextprotocol/inspector # point at http://localhost:8787/mcp
pnpm run typecheck # tsc --noEmit
pnpm test # vitest: store/simperium logic (in-memory, no network)
Layout
src/index.ts— OAuthProvider wiring (the entrypoint)src/mcp.ts—McpAgentDurable Object + the four toolssrc/store.ts— file↔note mapping, tag scoping (port ofstore.py)src/simperium.ts— Simperium HTTP client (port ofsimplenote.py)src/auth.ts— single-user OAuth login gateBOOTSTRAP.md— the doc you add to the Claude projectpython/— optional local stdio variant for Claude Desktop / Claude Code
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。