Roomie runs an MCP server. Claude, OpenAI, Kimi, Antigravity — if your AI speaks MCP, it can read your tasks, messages, emails, and calendar. No plugins. No API keys to manage.
We got tired of copy-pasting task descriptions into AI chats. So we built a way for the AI to just... look it up itself.
Run in your terminal:
claude mcp add --transport http roomie-workspace \
https://app.roomie.work/v1/api/mcpAuthorize with your Roomie account when prompted. That's it.
MCP stands for Model Context Protocol. Think of it as a universal plug between your AI and the apps you use. Instead of the AI being stuck reading only what you paste into the chat, MCP lets it query live data from your workspace on its own — your tasks, messages, emails, and calendar.
Anthropic introduced MCP in late 2024, and every major AI tool now supports it: Claude Code, Claude Desktop, OpenAI's Agent SDK, Kimi, Antigravity. It's an open standard, and the list keeps growing.
The practical upside: when you ask Claude “what's blocking the Acme deal?”, it doesn't need you to paste tasks or emails. It opens the right tool on its own and answers. You spend less time gathering context for your AI.
Once connected, the AI can read and search across everything in your workspace. You ask “what happened with the Acme deal this week?” and it actually looks through your messages, tasks, and emails to answer. No manual context needed.
These are real things you can say to Claude (or Kimi, or whatever you use) once Roomie is connected.
"Plan the implementation for ROOMIE-167"Claude reads the task, finds 3 related messages in #engineering, checks an email from the client, and writes a plan. You skipped 15 minutes of context gathering.
"Create a task from the last support email"The AI finds the email, creates a ticket on your board, assigns it, sets priority from the tone, and links the original thread. Took about 3 seconds.
"Summarize #product this week"Scans 47 messages, 12 task updates, and 3 calendar events. Gives you a summary you can paste into your standup doc. Actually useful on Monday mornings.
"Brief me on Acme Corp before my 2pm"Pulls the last 2 weeks of messages mentioning Acme, open tasks, and the contract email thread. You walk in knowing what happened instead of pretending you remember.
MCP is different because the AI lives where you already work — your terminal, your IDE, Claude Desktop — and Roomie is just one of its data sources. You pick the model. You pick the tool. Roomie gives it the data. Here's how it stacks up.
Automations are deterministic — you pre-define triggers and actions. MCP is conversational — the AI decides what to query based on your question. No workflows to design; no triggers that stop firing when you rename a field.
Those only work inside their own app, with the vendor's model, on the vendor's terms. MCP works in your terminal, your IDE, or any client — and swaps models without vendor lock-in.
A custom build means maintaining the integration, the auth, and the tool-schema glue for every model you support. MCP is a single open protocol — the server advertises its tools, and every compliant client uses them. Ship once, support every model.
Three short guides — one per client. Before you start, sign in to Roomie so OAuth has something to authorize against.
If you have Claude Code installed, this is a one-liner. Run the command below in any terminal:
claude mcp add --transport http roomie-workspace \
https://app.roomie.work/v1/api/mcpAuthorize with your Roomie account when prompted. To verify, ask Claude: “list my open Roomie tasks.” If it returns a real list, you're connected. To remove the server later, claude mcp remove roomie-workspace.
Open your Claude Desktop config file and add a mcpServers entry. On macOS the file lives at ~/Library/Application Support/Claude/claude_desktop_config.json. On Windows it's %APPDATA%\Claude\claude_desktop_config.json.
{
"mcpServers": {
"roomie-workspace": {
"command": "npx",
"args": ["mcp-remote",
"https://app.roomie.work/v1/api/mcp"]
}
}
}Restart Claude Desktop after saving. A browser tab opens for OAuth — sign in with your Roomie account and return to the app. You'll see “roomie-workspace” appear as an available tool at the bottom of the chat.
For Kimi, the CLI is similar to Claude Code:
kimi mcp add --transport http --auth oauth \
roomie-workspace https://app.roomie.work/v1/api/mcpFor OpenAI's Agent SDK or any generic MCP client, drop the URL into your server config:
{
"servers": {
"roomie-workspace": {
"type": "url",
"url": "https://app.roomie.work/v1/api/mcp"
}
}
}The server advertises its tools on connect, so the agent discovers what it can do with Roomie automatically — no manual tool-schema wiring on your side.
app.roomie.work and try again.MCP stands for Model Context Protocol. It's an open standard introduced by Anthropic in late 2024 that lets an AI agent read live data from an external app on its own, instead of relying on what you paste into the chat. Roomie runs an MCP server, so any MCP-compatible AI can query your tasks, messages, emails, and calendar directly.
Anything that speaks MCP. Today that includes Claude Code, Claude Desktop, OpenAI's Agent SDK, Kimi, and Antigravity. Generic JSON config works for anything else. When new MCP clients ship, they'll work with Roomie automatically — no update on your side.
No. For Claude Code and Kimi it's a single terminal command. For Claude Desktop it's pasting a short JSON block into your config file. We walk you through both on the setup page. Most people are connected in under two minutes.
The AI only sees what you, the logged-in user, can see. Authentication happens over OAuth with your Roomie account — no API keys to manage or leak. The AI cannot send messages, delete data, or act on your behalf unless you explicitly ask it to and approve the action.
Zapier runs deterministic workflows — you define triggers and actions in advance. MCP is the opposite: the AI decides what to look up, when, based on your question. No automations to design. You ask something, the AI figures out which Roomie tool to call and does it.
No. MCP is model-agnostic. Roomie works equally well with Claude, OpenAI, Kimi, Antigravity, and any future model that supports the protocol. You pick the model based on cost, speed, or capability — Roomie just provides the data.
MCP is included in the Team plan ($2 per user per month). No add-on fee, no separate subscription. The Free plan (up to 3 users) does not include MCP — it's meant for teams evaluating the core workspace before connecting AI.
Yes. The AI respects the same permission model as the rest of Roomie — private channels, DMs, and restricted projects are invisible to it unless the authenticated user has access. You can also revoke an MCP connection from the settings page at any time, which immediately cuts off the agent.
The Roomie MCP server runs on our infrastructure at app.roomie.work. It's not self-hostable today. If that's a hard requirement for your team, email us — it's on the roadmap for larger workspaces.
After authorizing, ask your AI something like "summarize my open tasks" or "what did the team discuss in #product yesterday?" If it returns real data from your workspace, you're connected. If it asks you to paste things in, it didn't pick up the server — check the setup config and re-run the command.
Still stuck? Send us a message — we reply within a day.
Connect your AI to Roomie. It can look things up on its own.
Free for up to 3 users. MCP on Team plan at $2/user/month.