prospective-memory
An MCP server that lets users capture quick phone thoughts via Android and later query them from AI assistants like Grok or Claude, functioning as a simple task memory store without notifications or geolocation.
README
Prospective Memory
Phone-first capture. MCP pull. Not Takeout.
Dump a thought in a few seconds on Android. Later ask Grok/Claude: “what’s open?”
No notifications in v1. No geofence. No Google history.
Sibling of google-activity-assistant — same kind of MCP host, different store.
v1 loop
Android (one box) --POST /v1/capture--> PC FastAPI + SQLite
│
Grok / Claude --MCP list_open_tasks----------┘
Server (Windows)
cd $env:USERPROFILE\Desktop\prospective-memory
uv sync
uv run pmem token # print X-PMEM-TOKEN for the phone
uv run pmem serve # http://0.0.0.0:8790
Leave serve running. In another terminal / Grok, MCP serve-mcp is already wired if you add the config below.
Phone must reach this machine:
- Same Wi‑Fi:
http://YOUR_LAN_IP:8790 - Out of house: Cloudflare Tunnel / Tailscale (see
docs/V1.md)
Android
- Open
android/in Android Studio (orgradlew.bat assembleDebug). - Set server URL + token on first launch (same token as
pmem token). - Save URL + token. Add lingo lines (
d=dahi lena). - Enable floating + tile (overlay permission) — tap
+→ type → Enter. Drag the+onto the bottom ✕ to hide it; Enable turns it back on. - Home screen widget (stretch wide for a type-here bar) or long-press app icon → Capture. Keyboard opens immediately.
- Empty Enter repeats the last lingo chip. Toast shows
category · text. Offline posts queue and flush when the network is back. - Optional Write (handwriting test) is in
android/.../ink/. First open downloads an on-device model. Not on the bubble. Drop it viaink/REMOVE.txt.
Allow HTTP cleartext for LAN IPs (already in network security config).
MCP (Grok)
~/.grok/config.toml:
[mcp_servers.prospective_memory]
command = "C:\\Users\\bda99\\.local\\bin\\uv.exe"
args = [
"run",
"--directory",
"C:\\Users\\bda99\\Desktop\\prospective-memory",
"pmem",
"serve-mcp",
]
enabled = true
startup_timeout_sec = 60
Ask: “list open tasks” / “anything grocery?” / “mark that dahi task done.”
CLI
uv run pmem capture "dahi lena"
uv run pmem list
uv run pmem stats
Non-goals (v1)
- OS reminders, geofence, WhatsApp check-in, glyphs, voice
- Merging with Google Takeout activity
- Multi-user cloud
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。