Agent or MCP?
When to use the built-in DataGlass Agent and when to connect DataGlass to ChatGPT or Claude through MCP.
Both use the same DataGlass tools and data. What differs is where you work and how much runs on its own. The Agent lives inside DataGlass and can run jobs on a schedule. The MCP connector brings DataGlass into the ChatGPT or Claude you already use.
| DataGlass Agent | MCP connector (ChatGPT / Claude) | |
|---|---|---|
| Where | Inside DataGlass: Agents, or Ask DataGlass on any page | Your ChatGPT or Claude account |
| Setup | None | Add the connector once and sign in |
| Sees the page you're on | Yes | No. Use the Ask AI buttons to send the on-screen context |
| Plans with targets & limits | Yes, approved once and then run step by step | No |
| Scheduled tasks | Yes, daily or weekly, reporting back on their own | No |
| Approving changes | Approve in DataGlass, or pick Bypass approval for a chat | Grant a write scope per domain, then your assistant asks before each change |
| Memory and skills | DataGlass memory and skills | Your assistant's own memory, plus the DataGlass playbooks |
| Mix with other tools | DataGlass only | Yes, e.g. your email, docs, or spreadsheets in the same chat |
| Plan | Pro, Enterprise, trial | Every plan. Pro features (Actions, ads optimization) still need Pro |
Pick the Agent if…
- You want work done on a schedule without opening a chat (Operator, KAM, margin guard).
- You want a plan with limits the agent is held to, such as a campaign budget or a margin floor.
- You'd rather not set anything up.
Pick MCP if…
- You already work in ChatGPT or Claude and want DataGlass next to your other tools.
- You want a one-off analysis to combine with files or apps outside DataGlass.
Many sellers use both: the Agent runs the shop day to day, and MCP answers questions inside the assistant they already use.