Security

Principles

1. Least privilege. Tokens carry scopes. The API enforces them on every route; routes without a scope are closed to tokens by default. The MCP server only lists the tools your token’s scopes allow.
2. No ambient authority. The MCP server has no database access and no rights of its own. It forwards your token to the API.
3. Humans confirm what is final or legal. Publishing, the rights statement, deleting, server keys and money stay in the portal.
4. No secrets in the conversation. Tool results never contain tokens, server keys or email addresses. Conversations are stored by AI providers, so anything in them counts as shared.
5. Untrusted content stays data. Text written by other users arrives inside an <untrusted-data> block with a fixed warning, stripped of control, invisible and direction-changing characters, and shortened. The content cannot close the block. Tool descriptions are fixed in code.
6. No fetching of arbitrary URLs. No tool loads a URL you or someone else provides.

Limits

Limit	Value
API calls per token	120 per minute, 2,000 per hour
MCP messages per token	300 per minute
Requests per IP without login	60 per minute
Request size	2 MB
Files per check_resource call	200 files, 1 MB in total

Audit log

Every API call a tool makes with your token is logged: time, token, action, result, client (MCP plus the name your client sends), tool and IP address. When an assistant connects, a Connected entry appears (at most every 10 minutes). Arguments and contents are not stored. You see the log in the portal under Account → AI activity. Entries are kept for 90 days.

Tokens

• Format gpat_ followed by 43 random characters. Stored only as a SHA-256 hash.
• Expire after 30 days by default, at most one year.
• At most 20 active tokens per account.
• Revoke instantly in the portal.

Reporting a vulnerability

Email security@gampeon.com. See Security reports.