Your AI assistant can now run your tchop app. Here’s how.

You open Claude. You type: “Post a push notification to all members about the town hall next Thursday at 3pm.”

Done.

No opening the tchop dashboard. No navigating to the push notification form. No copy-pasting. The AI called the right API, with the right parameters, and sent it.

That’s the tchop MCP Server. This is the practical guide to what it can actually do, how to set it up, and where its limits are.

What is an MCP server, and why does it matter here?

MCP (Model Context Protocol) is an open standard that lets AI assistants talk directly to external tools and APIs. Instead of you copying content from one place and pasting it somewhere else, the AI does it by calling the tool directly.

For tchop users, this means your AI assistant (Claude, Cursor, Windsurf) can create cards, send push notifications, moderate comments, and pull analytics from your tchop app — using plain language instructions.

The MCP server runs remotely at mcp.tchop.live. Nothing to install locally.

What you can actually do with it

Here’s the full scope of what’s live in v1 (Beta):

Content creation and publishing

• Create article, poll, image, video, and audio cards
• Publish, schedule, pin, or repost cards across mixes and channels
• Update existing cards

Push notifications and messaging

• Send push notifications to all members or specific segments
• Send in-app messages

Community moderation

• Read, hide, highlight, and reply to comments
• Manage tags, mixes, and channels

Analytics

• Query engagement data for cards and channels
• Pull analytics reports on demand

Real prompt examples that work:

“Create a poll card in the #internal channel asking the team to vote on the Q3 offsite location. Add three options: Berlin, Barcelona, Vienna. Publish it now.”

“Show me the five most-commented cards from the last 30 days.”

“Send a push notification to all members: ‘New employee handbook available. Check the Resources channel.'”

“Hide the flagged comment on the town hall article. Then reply: ‘Thanks for your feedback. We’ll follow up directly.'”

These aren’t hypotheticals. Each maps directly to a named tool in the tchop MCP. Any MCP-compatible AI assistant can call them.

What you can build on top of it

The more interesting use case isn’t one-off commands. It’s building workflows.

Automated content curator
Connect your AI assistant to an RSS feed or content source and have it draft and publish curated article cards on a schedule. You review before publish, or you don’t.

Moderation assistant
Set up a recurring task in Claude: check for new comments every 24 hours, flag anything below a sentiment threshold, draft replies for your review.

Analytics briefing
Every Monday morning, your AI pulls the week’s engagement data and sends a summary to your Slack or email. No dashboard needed.

Newsroom publishing workflow
Editors paste a story into Claude. The AI formats it, creates the card with the right tags, assigns it to the correct channel, and schedules it based on your publishing calendar.

Limitations worth knowing before you start

It’s Beta. The core toolset is solid, but it’s v1. The GraphQL API underneath is fully stable; the MCP layer is still evolving. Bug fixes are applied instantly — no reconnect needed. New or removed tools do require a reconnect.

Permissions are respected. The MCP can only do what your connected user account is allowed to do. If your account can’t delete a channel, the AI can’t either. This is by design.

AI output quality depends on your prompts. “Write something about the offsite” produces something generic. “Write a 100-word announcement card for the Q3 offsite on September 12-13 in Berlin, casual tone, aimed at the whole company” produces something usable.

One auth token equals one permission set. Different access levels require separate credentials. Get credentials from support@tchop.io — no self-serve yet.

Setup: how to connect in under 10 minutes

You need three things from tchop: your org URL (e.g. acme.tchop.io), an auth token, and an API client ID. Request them at support@tchop.io. Treat the auth token like a password — don’t paste it into shared Slack channels or screenshots.

Claude Desktop: Settings → Connectors → Add custom connector → set remote MCP server URL to https://mcp.tchop.live → enter org URL and auth token → authorize.

Claude Code (terminal):

claude mcp add --transport http tchop https://mcp.tchop.live   --header "x-tchop-org-url: acme.tchop.io"   --header "x-tchop-auth-token: YOUR_TOKEN"   --header "x-tchop-api-client-id: YOUR_CLIENT_ID"

Cursor or Windsurf: Add a tchop entry to your MCP config file pointing to https://mcp.tchop.live with the same three headers.

Full step-by-step guides for each client: api.tchop.live.

Who should use this now

Comms leads and community managers who already use Claude or another AI assistant: the MCP cuts the time between “I need to send something” and “it’s sent” to seconds.

Newsroom editors publishing to tchop-powered apps: have the AI format and schedule cards while you focus on the story.

Founders and lean teams running branded communities: build light automation that would otherwise require a developer.

Technical users who want to go deeper: the same GraphQL API powers the MCP. Build custom agents, deeper integrations, or automated pipelines with the same credentials.

One more thing

The MCP Server is the connective layer between your AI tools and everything tchop can do. Right now it covers the core — content, notifications, moderation, analytics. More operations are coming.

If you build something with it, tell us. The most useful production patterns shape what gets prioritized in v2.

Questions: support@tchop.io. Full API docs and operation reference: api.tchop.live.

Frequently asked questions about the tchop MCP Server

Do I need to be a developer to use the tchop MCP Server?

No. Setup requires pasting three credentials into your AI client settings — no coding involved. Once connected, you manage everything through plain language. Developers can also access the same operations via tchop’s GraphQL API.

Which AI assistants work with the tchop MCP Server?

Any MCP-compatible client. This currently includes Claude Desktop, Claude Code, Cursor, and Windsurf. As MCP adoption grows, the server will be compatible with new clients without changes on your side.

Is my data safe when using the MCP?

Yes. The MCP runs through tchop’s existing permission model — the AI assistant can only perform actions your connected account is authorized to perform. Your data stays inside tchop. The auth token functions like a password and should be kept private.

What’s the difference between the MCP Server and the tchop GraphQL API?

Both expose the same operations and share the same credentials and permission model. The MCP Server is designed for AI assistants via natural language. The GraphQL API is designed for developers building custom integrations programmatically.

The MCP Server is in Beta — what does that mean in practice?

The core toolset is stable and production-usable. The GraphQL API underneath is fully stable. Beta means the MCP layer may see adjustments. Bug fixes are applied immediately. When tools are added or removed, you’ll need to reconnect your client connector.