Security and limits
What an AI assistant can reach through Kalpana MCP, how to revoke it, and the limits it works within.
What an assistant can reach
- Only the one workspace you choose. When you connect an assistant, the consent screen asks which workspace to share, like an API key. It cannot see or change any other workspace, even ones you belong to. To use another workspace, reconnect and choose it.
- Only what your role allows. The assistant acts as you. If your workspace role cannot create or run batches, neither can the assistant.
- Only the permissions you approve. The consent screen lists them: seeing templates, images, batches and finished designs; creating batches and starting renders; adding images to your library; and, for the designer, saving new templates and designing and changing templates.
- No template internals on the batch server. Its tools return template names, previews and the inputs a batch needs, never the underlying design data. Kalpana Designer reads and edits layers, because designing them is its job; it is a separate connection, and only roles that can edit templates can use it.
Credits stay under your control
Creating a batch does not render it. Before a batch runs, the assistant must show you its exact credit cost and get your confirmation. Every change it makes, such as creating or editing a batch, also needs your approval first. You can also ask it to stop a running batch: cancelled jobs are not charged. See Kalpana Batch.
Designer calls are different: they use the workspace's render credits as the assistant works, one credit for every 20 calls that make or draw a design. Reads, checks, helpers and housekeeping are free. See Credits. Plans and prices are on the pricing page.
Revoke access
Open Settings → Connected assistants in Kalpana and choose Disconnect on the assistant's Batch MCP or Designer MCP entry. An assistant connected to both servers has one entry for each. Access ends straight away. To use the assistant again, connect it again from the assistant.
Kalpana never writes your sign-in tokens to its logs.
Rate limits
Limits apply per minute. When one is reached, Kalpana answers with 429 and a Retry-After: 60 header; wait a minute and try again.
| Activity | Limit |
|---|---|
| Batch tool calls from one connected assistant | 60 per minute |
| Designer tool calls from one connected assistant | 60 per minute, counted separately |
| Sign-in and token requests | 10 per minute |
| MCP requests from one network address | 300 per minute |
Normal use stays well inside these limits. An assistant that loops on a tool, or a script that retries without waiting, is the usual cause of a 429.
Request size
| Request | Largest size |
|---|---|
| A tool call | 1 MB |
| A sign-in request | 16 KB |
Larger requests are refused with 400 or 413, and a tool call must be sent as application/json. For a big batch, create it in parts and add rows with update_batch.
Found a security problem? Email support@kalpana.one.
Updated 4 Nov 2026