Integrations gateway for agents with 2FA for destructive ops (OSS)

Free

FreeFree tier

Type

Open Source

About Integrations gateway for agents with 2FA for destructive ops (OSS)

AgentPort is an open source gateway designed to securely connect autonomous AI agents to external services. It provides a security layer that keeps API keys hidden from agents and gives users granular control over what tools agents can use and when. AgentPort ships with over 50 pre-built integrations including Stripe, PostHog, GitHub, Gmail, and Google Calendar. Users can connect an integration with a few clicks and immediately make it available to agents via CLI or MCP. For each tool, administrators can set approval policies: auto-approve (agent runs the tool automatically), ask for approval (agent returns a link for manual approval of exact parameters), or deny (agent cannot use the tool). This enables deploying autonomous agents securely while giving them access to sensitive operations like creating refunds or sending emails.

Key Features

Open source gateway for connecting agents to external services

Over 50 built-in integrations (Stripe, PostHog, GitHub, Gmail, Google Calendar, etc.)

Granular approval policies for each tool: auto-approve, ask for approval, or deny

Agents never see API keys – keys remain secure on the gateway

Connect agents via CLI or MCP (Model Context Protocol)

Self-hostable with Docker Compose or production install script

Cloud-hosted option available at app.agentport.sh

Pros & Cons

Pros

Open source and free to self-host
No API key exposure – keys stay on the gateway
Fine-grained control over which tools agents can use, including parameter-level approval
Easy setup with many pre-built integrations
Supports both CLI and MCP for agent connectivity
Can be deployed locally or in the cloud

Cons

Requires setup and maintenance of a self-hosted instance
Approval workflow may slow down fully autonomous execution
Limited to CLI and MCP integration methods (no direct API/HTTP interface for agents)
Relatively new project with a smaller community and fewer integrations compared to established platforms

Best For

Give an AI agent access to Stripe to manage customers and refunds with manual approval for destructive actionsLet an agent read and compose emails via Gmail while requiring user approval before sendingConnect an agent to GitHub to search issues and create pull requests under controlled permissionsEnable agents to interact with multiple services (calendar, email, CRM) through a single secure gateway

FAQ

What is AgentPort?

AgentPort is an open source gateway that securely connects AI agents to external services. It acts as a security layer that hides API keys from agents and lets you control exactly what tools they can use and when.

How do approval policies work?

Each tool can be set to one of three policies: Auto-approve (agent runs the tool without user intervention), Ask for approval (agent returns a link that only the logged-in user can approve, with exact parameters), or Deny (agent cannot use that tool at all).

What integrations are available?

AgentPort ships with over 50 integrations out of the box, including Stripe, PostHog, GitHub, Gmail, and Google Calendar. More are being added.

How do agents connect to AgentPort?

Agents can connect to AgentPort using either the provided CLI or the MCP (Model Context Protocol) interface.