Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@DayLoop MCP Server I'm 40 minutes behind — replan my day and keep my client promises"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Simulated experience. Built for the Alexa+ track of the Amazon Developer Hackathon, simulated experience path. There is no real Alexa skill, calendar or inbox, and all data is invented. The plans, promise checks, drafts and memory are computed from that data by the code in this repo, and the same tools are served by a real MCP server over Streamable HTTP.
A calendar manages time. What a freelancer actually loses clients over is a broken promise: "you'll have the proposal today", "the invoice goes out tonight", "the bug is fixed before our 4 o'clock". When a meeting runs long, those promises break quietly, and the client is the first to notice.
DayLoop keeps a ledger of those promises. When the day slips, it works out which promise is about to break and drafts an honest heads-up before the client has to chase you. Nothing leaves DayLoop without your approval.
Related MCP server: Nexxupp MCP Server
Every step is a tool call, listed live in the Agent steps panel on the right.
The numbers are not scripted. Say you are 20 or 90 minutes behind, or ask for "no calls before 11", and you get a different plan, a different draft and a different Day 2 (see test/dayloop.test.js).
No install needed: https://zroit.github.io/dayloop/ runs the whole agent in your browser. Press Reset demo, then tap the suggested line under the conversation each time.
Locally:
npm install
npm run dev # http://localhost:5173, agent in the browser, state in localStorage
npm test # planner, promises, intents and both days end to endnpm run mcp # DayLoop MCP server at http://localhost:8787/mcp (Streamable HTTP)Web app on top of it: open http://localhost:5173/?mcp=http://localhost:8787/mcp. The header shows MCP: localhost:8787 and every turn runs in the server.
Any MCP client: connect to http://localhost:8787/mcp, for example with npx @modelcontextprotocol/inspector.
Scripted client: npm run mcp:demo plays both days over MCP and prints every step.
State lives in mcp/.dayloop-state.json, so memory survives a server restart the way it survives a browser reload.
web app (React) any MCP client
conversation | Echo Show cards | steps (Alexa+ agent, Claude, Inspector)
| |
local backend ----- or ----- MCP client ---- Streamable HTTP
| |
+------------------+ +-------------------+
| |
src/core/agent.js one turn: parse -> call tools -> decide -> queue for approval
|
src/core/tools.js one registry, 19 tools (+4 agent tools in the server)
+-----------+-----------+---+-------+-------------+------------------+
calendar tasks team chat client inbox memory + clock planner + promisesPath | What it does |
| The simulated adapters. Each adapter is its tools: name, description, zod input schema, MCP annotations ( |
| Conflict resolution (the lower-ranked meeting moves), learned preferences ("no calls before X"), earliest-fit packing by priority, and |
| Rule-based understanding of what the user says: "half an hour", "done except the report", "no meetings until 11:30am". Deterministic on purpose, so a demo and a test give the same answer. |
| The orchestration. It never touches state directly; it calls tools and records each call as a step. Messages and calendar moves wait in a pending approval, and |
| A stateless Streamable HTTP MCP server on |
| The Echo Show style UI: timeline cards, the promise ledger, the approval tray and the live steps panel. |
Real | Simulated |
Planner, promise ledger, replanning, drafts and memory, all computed from state | Calendar, tasks, team chat and inbox contents ( |
MCP server over Streamable HTTP, usable by any MCP client | No Alexa device; the browser stands in for the Echo Show screen |
Approval guardrail: nothing is sent without your yes | "Sent" messages go to a simulated outbox |
State that survives reloads and server restarts | Day 2 is reached with a button, not by waiting a night |
Built with Claude Code.
MIT, see LICENSE. Made by ZroIT.
This server cannot be deployed
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
Nifty's MCP server — exposes tasks, projects, messages, and files as tools for AI agents.
Persistent memory, knowledge graph, task management and more for AI agents via MCP.
Hosted MCP memory and agent control plane for durable conversations, jobs, and operations.
