Connect Claude or ChatGPT to Zentesimal with MCP

Connect an AI assistant like Claude or ChatGPT to Zentesimal over MCP, and it can manage Workers, Skills, Actions, Guards, and Dynamic columns, or search Voice of Customer data — directly from a chat, with no manual API key.

Copy your MCP server URL

1. In Zentesimal, go to Account → MCP.

1. Click Copy URL. You'll paste this into Claude or ChatGPT in the next step.

Connect Claude

1. In Claude (claude.ai or Claude Desktop), open your connector settings and add a new custom connector.
2. Paste the server URL you copied from Account → MCP.
3. Claude opens a browser window and asks you to log in to Zentesimal and approve the requested permissions.

1. Once approved, Claude can call your Workers, Skills, Actions, Guards, fields, Dynamic columns, Results, and Voice of Customer tools from chat.

Connect ChatGPT

1. Ask a ChatGPT workspace admin to turn on Developer mode. On ChatGPT Business or Enterprise, it's in the workspace admin settings under apps/connectors. Custom apps need it until Zentesimal is listed in the ChatGPT app directory.
2. In ChatGPT, open Settings → Apps, turn on Developer mode for your account, then click Create app.
3. Name it Zentesimal, paste the server URL you copied from Account → MCP (for example https://app.zentesimal.com/api/mcp), and choose OAuth as the authentication method.
4. ChatGPT opens a browser window and asks you to log in to Zentesimal and approve the requested permissions.
5. Once approved, add the Zentesimal app to a conversation and ask it to list your Workers. ChatGPT asks you to confirm before running tools that change or archive something.

Coming soon: connect from the ChatGPT app directory

Zentesimal has been submitted to the ChatGPT app directory. Once it's listed, you can find Zentesimal in ChatGPT's Apps directory and click Connect, with no Developer mode or custom app setup. The login and permission steps stay the same.

What the connector can do

• Workers — create, rename, update, duplicate, or archive a Worker
• Knowledge sources — attach knowledge sources to a Worker
• Input and output fields — add or update Input/Output Settings extras on a custom Worker or a built-in one (for example Chat). These are not Dynamic columns.
• Skills and Guards — add, update, enable/disable, or archive them on a Worker. When you create a Guard, write the criterion from the Worker's final result only — the Guard does not see conversation history, knowledge, Skills, or tool traces.
• Dynamic columns — list, add, update, or run Prompt, Keyword matching, or Classification columns. Ask the assistant to list columns first so @field names are valid. New cells fill after a Result is processed or failed. To fill past Results, ask it to run the column on a date range — the same as Results → Run column.
• Run a column on past Results — fill or refresh one Dynamic column for a date range, the same as Results → Run column. Up to 5,000 Results per run. It overwrites existing cells in that column, so the assistant asks you to confirm first.
• Actions — list the Actions you can add, add one to a Worker so every Skill inherits it, or add it to a single Skill only. Change settings, override or turn off an Action per Skill, or remove it.
• Results — list a Worker's results, or group them by a native field or Dynamic column for counts. Grouping on a Dynamic column uses only that column (one value per Result).
• Voice of Customer — keyword overview, semantic search over indexed results, and anomaly detection
• Internal Q&A — ask the Internal Q&A Worker a question (the same Skills and knowledge it uses in Slack, without posting to Slack)

Nothing is permanently deleted. Archiving a Worker or Skill is reversible, and Results are kept. Tools that only read data run without a confirmation prompt; tools that overwrite or archive content ask you first.

Add and configure worker Actions from chat

Actions are what a Worker can do besides answering, like sending a Slack message or updating a Crisp conversation. With Create and edit access, the assistant manages them at two levels:

1. Ask the assistant which Actions it can add. It lists the same catalog you see in the app's action picker, including the settings each one needs.
2. Tell it where to add the Action — the Worker pool or a single Skill — and give it the settings, like the Slack channel.
3. To change an inherited Action for one Skill, ask it to override the settings or turn the Action off on that Skill. The pool Action stays on for every other Skill.
4. Open the Worker in Zentesimal and go to Knowledge → Agent skills. Worker pool Actions appear in the Actions row, and each Skill shows its own Action icons.

If a tool is not allowed

When you connect, Zentesimal asks you to grant View and/or Create and edit. If you only granted View, the assistant can still try a create or edit tool, but Zentesimal returns a permission message instead of failing as if the connector broke. Ask the assistant to reconnect and allow Create and edit Workers, Skills, Actions, and Guards.

Internal Q&A teammates

If someone was invited with the Internal Q&A role, they log in and land on Account → MCP. After they approve the connector, their assistant can only ask Internal Q&A. It may still list Worker tools, but using them returns a permission message — reconnecting will not unlock Workers, Skills, or Results. They should still paste the same /api/mcp URL from that page.