zenture
Get startedLog In

Available

Evaluate inside the AI environments you already use.

Plugin and workflow integrations bring the reliability layer closer to ChatGPT, Claude and other AI-assisted work without replacing the user’s context.

See the evaluation engine

ChatGPT integration

Connect zenture to ChatGPT.

zenture is a published plugin, available on every plan including Free. Install it once and ask it to check any answer directly in your chat.

Setup in five steps

  1. Open Plugins

    In ChatGPT, go to Settings > Plugins. No developer mode needed — on Enterprise or Edu plans, an admin may need to turn plugins on first.

  2. Find zenture

    Search the plugin directory for zenture and open its listing.

  3. Install and connect

    Choose Install plugin, then Connect.

  4. Sign in and approve

    A browser window opens. Sign in to zenture and approve the requested access.

  5. Use it in a chat

    In a chat, mention @zenture or pick it from the + menu, and ask it to check an answer. Approve the request when ChatGPT asks. Disconnect it later from Settings > Plugins.

Claude integration

Connect zenture to Claude.

Add zenture as a custom connector in Claude, then ask it to check an answer without leaving your conversation.

Setup in five steps

  1. Open Connectors

    In Claude, open Customize, then Connectors.

  2. Add the connector

    Select +, choose Add custom connector, and paste https://mcp.zenture.app. Free plans allow one custom connector; on Team or Enterprise, an owner adds it once under Organization settings > Connectors, then members can connect.

  3. Connect and approve

    Select Connect. A browser window opens — sign in to zenture and approve the requested access.

  4. Enable it in a chat

    Open the + menu in your conversation, choose Connectors, and turn zenture on.

  5. Check an answer

    Ask Claude to check an answer with zenture. The first time, allow the tool — Allow once or Allow always — and the result appears in the chat.

MCP / API

Use MCP or API.

Connect zenture as an MCP server for clients like Cursor or VS Code, or call the Public API from your own backend. The Public API is currently a release candidate.

Setup in five steps

  1. Install the client

    Install the zenture Python client from PyPI with pip install zenture.

    zenture-client
    $ pip install zenture

  2. Create access

    Create an API token in the zenture web app under Profile > API tokens for the client or API, or sign in with OAuth when your MCP client adds the server.

  3. Connect

    Add https://mcp.zenture.app as a remote MCP server in your MCP client, or set the ZENTURE_API_KEY environment variable for the client or API.

    mcp.json
    {
      "mcpServers": {
        "zenture": {
          "url": "https://mcp.zenture.app"
        }
      }
    }

  4. Start the first Run

    Call the run tool from your MCP client, call client.runs.run(...) with the Python client, or send POST /v1/runs/prepare and then POST /v1/runs with an Idempotency-Key header. The classic evaluate remains available as a secondary path.

    run.py
    from zenture import ZentureClient
    with ZentureClient.from_env() as client:
        run = client.runs.run(
            task="Check that the answer names exactly two colours.",
            artifact={"type": "text", "value": "Blue and green."},
            idempotency_key="review-case-123",
        )
        client.runs.wait(run.run_id, timeout=120.0)
        result = client.runs.get(run.run_id, view="full")
        print(result.status, result.acceptance_decision)

  5. Read the result

    Through MCP, get_run returns the status, acceptance_decision, result content, and next action; list_runs, cancel_run, and record_run_outcome cover the rest of the Run lifecycle. Through the API, read GET /v1/runs/{id}.