MCP connector

The MCP connector gives an AI assistant 51 tools: employees, fields, templates, departments, campaigns, deployments, analytics, team and activity log. It acts with the role of the person who authorized it: a member’s connector can’t change settings, brand or fields, bulk-delete employees, or invite. You describe what you want; the assistant calls the right tools on your workspace. From the Starter plan up.

Remote connector (recommended)

Nothing to install. The server address is:

https://app.gensy.email/api/mcp

• Claude.ai and Claude Desktop — Settings → Connectors → Add custom connector.
• ChatGPT — Settings → Connectors → Developer mode.
• Claude Code and Cursor — add the address as a remote MCP server.

Authorization uses OAuth: the consent screen opens in Gensy, where you choose the workspace and scope. The access appears in Active access with the “Connector” badge and can be revoked in one click; its token is valid for 30 days and renewed automatically.

Local server

For a client that doesn’t support remote connectors, npx -y gensy-mcp works with an API key. See the API keys page.

Load fewer tools

The AI assistants & API keys tab lets you check the toolsets you want; the address updates accordingly, for example …/api/mcp?toolsets=employees,analytics. Fewer tools means less context used up in the assistant.

The 51 tools, in ten toolsets

• workspace — overview, settings, brand.
• employees — list, read, create, update, bulk update, assign, duplicate, delete.
• templates — list, read, create, update, publish, delete, gallery.
• departments, campaigns, fields — full management.
• signatures — generate, send by email, deploy to Gmail or Outlook, unpublish, schedule.
• analytics — views and clicks.
• team — members and invitations.
• activity — log and alerts.

Example requests

• “How many employees don’t have a template assigned?”
• “Put the whole Marketing department on the bilingual template, then schedule the deployment for Monday at 9 a.m.”
• “Which links were clicked the most in the last 30 days?”

Security

• Destructive tools — deletions (employees, templates, departments, campaigns, fields), replacing fields, unpublishing, scheduling a removal, cancelling a schedule, removing a member — require confirmation (confirm: true): the assistant must ask you for it.
• A read-only scope exposes no write tools.
• Responses pass through the model provider (Anthropic, OpenAI…) and are subject to its terms; the consent screen reminds you of this. Portal tokens and billing identifiers never go through this channel.

Responses that are too long

Every tool result is cut at 120,000 characters ([truncated: …]). Ask the assistant to paginate (page, per_page: 50 by default, 200 at most), filter, or reduce the level of detail (detail: "minimal" or "summary"). MCP tools have no limit parameter.