kitchen
Deployment target for the server: it can be deployed to a Cloudflare account via the cf CLI (or the 'Deploy to Cloudflare' button), with the D1 schema applied automatically at runtime, secrets managed through Cloudflare Workers secrets, and optional custom domains attached from a Cloudflare zone.
The server is designed to be deployed and self-hosted as a remote MCP server on Cloudflare Workers with a D1 database binding, and ships with a Deploy to Cloudflare button plus a cf CLI deploy flow that builds the Worker, provisions the 'kitchen' D1 database, and uploads it.
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., "@kitchenadd the ingredients for tonight's lasagna to our grocery list"
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.
kitchen
A shared kitchen for your cooking assistants. kitchen is a small, self-hostable remote MCP server on Cloudflare Workers + D1. Each person in a household connects their own AI cook bot (built for Grok bots; any MCP client works). The bots then share:
Recipes: create, read, update (with conflict detection), delete, search.
One grocery list: add items, check them off, remove them, clear checked items, add a recipe's ingredients in one call.
Context notes: preferences, dietary needs, summaries of what someone discussed with their bot, meal plans. Each note records its author and timestamps.
Everything is an MCP tool. There's no web UI. Each member has their own revocable bearer token, and partners join with single-use invite codes.
The button has not yet been verified with this project's
cloudflare.config.ts(the project uses the newcfCLI rather than Wrangler). See open questions. The manualcf deploypath below is the reference.
Design notes (tool schemas, D1 schema, auth flow, trade-offs): docs/DESIGN.md.
How it works
Whoever deploys the server sets a secret,
KITCHEN_ADMIN_SECRET.Your bot connects to
https://<your-worker>/mcpwithout a token and callscreate_kitchenwith that secret. It gets back a token for you.Reconnect with
Authorization: Bearer <token>. Callinvite_member, which returns a code likeKTCH-7QX4-M9PB-2RZD(single use, 60 minutes by default).Your partner's bot connects without a token and calls
redeem_invitewith the code. It gets back its own token.Both bots now see the same recipes, grocery list, and notes.
Lost or leaked token? Call rotate_token. Remove someone? Call revoke_member.
Related MCP server: samskriti-project
Requirements
Node.js ≥ 22.18 and pnpm (npm works too)
The
cfCLI comes as a dev dependency:pnpm exec cf …, ornpx cf …
Local development
pnpm install
cp .dev.vars.example .dev.vars # set KITCHEN_ADMIN_SECRET to anything for local use
pnpm dev # cf dev → http://localhost:5173/mcpThe Worker creates its D1 tables on the first request, so there's no migration step. Local data lives in .cloudflare/state. Delete that folder to start over.
Run the end-to-end sample session against the dev server (create kitchen → invite → redeem → recipe → groceries → note → read back as the partner):
pnpm sample-session # or: node scripts/sample-session.ts http://localhost:5173/mcpExplore by hand with the MCP Inspector: npx @modelcontextprotocol/inspector, Streamable HTTP, URL http://localhost:5173/mcp, then add an Authorization: Bearer … header once you have a token.
Tests
pnpm test # Vitest 4 + @cloudflare/vitest-pool-workers: runs inside workerd with a real D1
pnpm typecheck # cf workers types && tscSelf-hosting with the cf CLI
pnpm install
pnpm exec cf auth login
pnpm run deploy # cf deploy: builds, provisions the "kitchen" D1 database, uploads
pnpm exec cf workers secrets update KITCHEN_ADMIN_SECRET --worker kitchen --text "$(openssl rand -base64 32)"Alternatively, put KITCHEN_ADMIN_SECRET=… in a file and pass it at deploy time with pnpm exec cf deploy --secrets-file <file>. Keep the secret somewhere safe: anyone who has it can create kitchens on your server.
The schema is applied by the Worker on its first request (src/migrate.ts), using the same d1_migrations bookkeeping as cf d1 migrations apply, so you don't need to look up the database ID.
Custom domain
The intended production deployment is https://kitchen.lesentiel.com/mcp. That domain is attached only in the lesentiel mode:
pnpm run deploy:lesentiel # cf deploy --mode lesentielFor your own domain, edit the domains line in cloudflare.config.ts. The zone must be on the account you deploy to. A plain pnpm run deploy stays on *.workers.dev.
Grok plugin
plugin/ is a Grok plugin: an MCP server config plus a skill that teaches Grok the setup flow and tools. Install it with grok plugin install lesentiel/kitchen#plugin --trust, then set KITCHEN_TOKEN (and KITCHEN_MCP_URL if you self-host). See plugin/README.md. Validate changes with grok plugin validate plugin.
Connecting a bot
Any MCP client that supports remote Streamable HTTP servers with custom headers:
URL:
https://<your-worker-or-domain>/mcpHeader:
Authorization: Bearer kit_…(omit it only forcreate_kitchen/redeem_invite)
The server speaks the 2026-07-28 MCP protocol and also serves 2025-era clients statelessly.
Project layout
cloudflare.config.ts cf config: D1 binding, secret, domain (generated by `cf init`, then extended)
migrations/ D1 schema (applied by the Worker at runtime; compatible with `cf d1 migrations`)
src/index.ts fetch entry: homepage at /, bearer auth, MCP handler at /mcp
site/index.html the one-page homepage (bundled into the Worker)
src/server.ts MCP tools and zod schemas
src/{kitchen,recipes,grocery,notes}.ts data access, always scoped to the caller's kitchen
test/ Workers-runtime tests
scripts/ sample session
plugin/ Grok plugin (manifest, .mcp.json, skill)
docs/DESIGN.md design and decisionsLicense
MIT
This server cannot be deployed
Maintenance
Related MCP Connectors
One memory, every AI. A shared, user-owned markdown memory your AI clients read and write over MCP.
Nifty's MCP server — exposes tasks, projects, messages, and files as tools for AI agents.
Cloud-hosted MCP server for durable AI memory
Remote MCP server for The Colony — a social network for AI agents (posts, DMs, search, marketplace).
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceMCP server that gives AI agents full control over grocery lists, todos, and packing lists. Your AI creates lists, adds items, checks them off, and shares with family/friends.4MIT
- AlicenseAqualityCmaintenanceA local MCP server that enables multiple AI coding tools to share structured project state (decisions, tasks, bugs) so they coordinate without re-explaining.5MIT
- FlicenseNot gradedqualityCmaintenanceA remote MCP server that gives AI assistants a shared memory, enabling structured handoffs, journaling, and notes for seamless continuation across sessions.-
- AlicenseNot gradedqualityFmaintenanceA self-hosted MCP server enabling multiple AI coding agents to share state, preserve context across sessions, and coordinate with each other.15 npmApache 2.0