AgentPort: MCP multi-agent credential manager

See and control which AI agents touch your SaaS

Problem

SaaS products are shipping MCP servers fast, and the same user account now connects Cursor, Claude, ChatGPT and other agents simultaneously. Operators report the biggest pain is not being able to tell agents apart: when one agent starts looping or hammering a search endpoint, teams cannot identify the source, rate-limit it, or revoke just that agent.

Solution

A lightweight proxy gateway that customers point their MCP server at. It issues separate credentials per AI client, tracks per-client call volume, error rates and loop patterns, and lets teams set rate limits, pause, or revoke access for a single agent. Anomalies trigger email/Slack alerts with a one-click block button.

Users

Small SaaS teams and indie developers who already run an MCP server; API product owners dealing with multi-agent abuse.

MVP

MCP reverse proxy with multi-tenant credential issuance; per-client call logs and usage stats; loop and high-frequency request detection rules; per-agent rate limit and revocation toggle; webhook alerts to Slack/email.

Revenue

Tiered subscription by number of agents or monthly call volume; free tier with 2 agents; team plan priced per seat.

Why now

MCP became the default SaaS entry point in 2026, but most teams shipped 'connectable' without 'controllable' — the exact gap gateway tools fill.