MCP Server

The Model Context Protocol is the open standard AI assistants use to call outside tools. It is not tied to one vendor: Claude, ChatGPT, Cursor, Gemini CLI, Zed, Windsurf and most agent frameworks speak it. ChatGenius offers it two ways. Hosted: paste https://sumgenius.ai/api/mcp into your assistant, click Connect, approve it once in ChatGenius. That is the path for claude.ai, ChatGPT, phones, and anything that cannot run a local program. Local: the chatgenius-mcp npm package runs on your computer with an API key, for Claude Desktop, Claude Code, Cursor and the like. Both speak MCP to the assistant on one side and call the REST API on the other, and every tool is exactly one API call.

Once it is installed you can ask things like "who messaged us on Instagram this week that we have not replied to", "book Maria for Thursday at 2pm and tell her", or "build a flow that answers price with our packages and collects an email", and the assistant does it with the same calls a developer would make.

npx -y chatgenius-mcp

Add https://sumgenius.ai/api/mcp wherever your assistant takes an MCP server or a custom connector. It discovers our sign-in on its own, registers itself, and sends you to a ChatGenius page where you choose what it may do: Read only, or Read and act (send, book, build, publish). You must be the account owner, and the account needs the Webhook & API add-on.

One click

Both hand the app the hosted server already filled in. You still sign in and choose what it may do.

That is OAuth 2.1 under the hood: authorization code with PKCE, public clients, one-hour access tokens with rotating 30-day refresh tokens, and no admin scope ever. We email the owner when an assistant is connected. Disconnect it any time in the portal under Settings, AI assistants, or from the assistant's side.

Where to paste it

• Claude

  Settings, Connectors, Add custom connector. Leave the detected defaults. Works on the web, desktop and phone.

• ChatGPT

  Settings, Connectors, Create, with developer mode switched on.

• Cursor

  Settings, MCP, Add new server, type Streamable HTTP.

• Claude Code

  claude mcp add --transport http chatgenius https://sumgenius.ai/api/mcp

Building your own app? The discovery documents are at /.well-known/oauth-authorization-server and /.well-known/oauth-protected-resource. Dynamic client registration and client ID metadata documents are both accepted. The OAuth section of API authentication has the details.

https://sumgenius.ai/api/mcp

claude mcp add --transport http chatgenius \
  https://sumgenius.ai/api/mcp

Authorization: Bearer sgwh_your_api_key_here

For Claude Desktop, Cursor, Zed, Windsurf and other clients that run local servers. You need Node 18 or newer and an API key. Create the key in the portal under Webhook & Send API, Credentials, and give it the scope you want the assistant to have: read for a look-but-do-not-touch assistant, write for one that can send, book and build.

Then add the server to your client. Claude Desktop reads claude_desktop_config.json; Cursor, Zed and Windsurf take the same command, arguments and environment in their own MCP settings. Claude Code has a plugin: two commands, and it asks for your key.

Environment

• CHATGENIUS_API_KEYrequired

  Your API key. Its scope caps what the tools can do; the API enforces that on every call.

• CHATGENIUS_MAX_SCOPEoptional

  read, write or admin. Hides every tool above that scope, so the assistant never sees them.

{
  "mcpServers": {
    "chatgenius": {
      "command": "npx",
      "args": ["-y", "chatgenius-mcp"],
      "env": { "CHATGENIUS_API_KEY": "sgwh_your_api_key_here" }
    }
  }
}

claude plugin marketplace add SumCodeAI/chatgenius-claude-plugin
claude plugin install chatgenius@sumgenius

105 tools in ten groups, each mapped to one endpoint of the REST API and carrying the same rules, limits and error codes. Read tools are safe to run freely. Write tools create or edit records. Destructive tools reach a real person, publish content or change what runs live; they are marked so your client asks you before running them.

Group	Tools
Account	get_account, get_overview
Contacts	list_contacts, get_contact, upsert_contact, update_contact, list_tags, create_tag, tag_contact, untag_contact, delete_tag, list_custom_fields, create_custom_field, update_custom_field
Inbox	list_conversations, get_conversation, list_messages, update_conversation, send_message, assign_conversation, list_team_members, list_blocked_users, block_user, unblock_user, list_opt_outs
Appointments	list_appointments, get_availability, get_appointment, update_appointment, book_appointment, reschedule_appointment, cancel_appointment
Flows	list_flows, get_flow, get_flow_definition, get_flow_node_catalogue, validate_flow, create_flow, update_flow, publish_flow, activate_flow, pause_flow, duplicate_flow, discard_flow_draft, delete_flow, get_flow_analytics, list_flow_sessions, get_flow_session, list_flow_session_events, list_flow_events
Comments	list_comment_triggers, get_comment_trigger, simulate_comment, create_comment_trigger, update_comment_trigger, duplicate_comment_trigger, delete_comment_trigger, get_comment_trigger_stats, list_recent_posts, list_handled_comments, get_handled_comment, get_comment_manager_stats, approve_comment_reply, reject_comment_reply, bulk_approve_comment_replies, bulk_reject_comment_replies
Content Studio	list_posts, get_post, get_posting_capabilities, create_post, update_post, schedule_post, publish_post, unschedule_post, retry_post, delete_post
Market Intel	list_competitors, get_competitor, get_competitor_leaderboard, get_follower_history, get_wins_and_opportunities, get_audience_demographics, get_top_posts, get_account_insights, add_competitor, remove_competitor, refresh_market_intel
Knowledge base	list_faqs, create_faq, update_faq, list_services, list_knowledge_gaps, get_knowledge_usage, get_faq, delete_faq, get_service, create_service, update_service, delete_service, list_documents, get_document, update_document, delete_document
AI settings	get_ai_config, get_ai_channels

Managing the team, webhook endpoints, API keys and media uploads, and switching the AI on or off, are not exposed: they need the admin scope, which an assistant never gets. Use the portal or the REST API for those.

• Scope is enforced twice

  The API checks the key's scope on every call. CHATGENIUS_MAX_SCOPE is a second, local cap that hides tools from the model altogether.

• No double sends

  Every write carries an Idempotency-Key, so a retried tool call cannot send a message or book a slot twice.

• Customer text is data

  Message and comment text in results was written by your customers. The server instructs the model to treat it as data, never as instructions. Keep destructive tools behind your client's approval prompt.

• Revoke any time

  Revoke the key in the portal and the server stops working on the next call.

https://www.npmjs.com/package/chatgenius-mcp