Connectors & triggers
Connectors overview
How RunAIAgents connects your agents to third-party services, and how those credentials are protected.
2 min read
A connector lets an agent call a real third-party service — read a Slack channel, create a Salesforce record, send an email through Google Workspace — without you writing any API integration code yourself.
What's available
RunAIAgents ships with 76+ first-party HTTP-API connectors covering services like Google Workspace, Microsoft Graph, Slack, Salesforce, Zendesk, HubSpot, and Notion. Beyond the first-party set, you get:
- Composio, giving you access to roughly 980 additional toolkits.
- A curated set of ~10 vetted remote MCP servers — Model Context Protocol servers whose tools an agent can call directly.
Between first-party connectors, Composio, and MCP, most services you'd want to integrate are already covered. If one isn't, api_call lets an agent hit any HTTP endpoint directly.
How an agent uses a connector
Once you've connected a service (see connecting a service), an agent calls it through the connector_action tool. You pick which connectors an agent can use when you build it — conversationally or on the canvas — and the model decides when to invoke them as part of its tool-calling loop, the same as any other tool.
MCP server tools work the same way: once an MCP server is added to an agent, its tools show up alongside your other connectors and the model can call them.
Credential security
Every credential you connect — an OAuth token, an API key, a paste-in secret — is encrypted with AES-256-GCM before it's stored. A few things follow from that:
- Credentials are org-owned: they belong to your organization, not an individual user, and are scoped to your tenant.
- Once saved, a credential is never shown back to you. If you need to verify what's connected, you see the connection's name and status, not the secret itself.
- The same encryption module is used everywhere secrets are stored — there's no separate, weaker path for connector credentials.
Per-tenant hosts
Some services you connect aren't at a fixed public URL — think an internal Jira instance or a self-hosted GitLab. RunAIAgents supports per-tenant hosts, so you can point a connector at your own instance's address instead of a shared default.
Self-healing connectors
Third-party APIs change: a field gets renamed, an endpoint gets deprecated. When a connector call breaks in a way the platform can diagnose, it can propose an automatic patch to the connector definition. That patch doesn't apply itself — it sits behind an approval gate, so a human reviews and approves the fix before it's used in a live agent.
Next steps
To start using a connector in an agent, see connecting a service for the step-by-step flow. To trigger a run automatically once your agent is wired up, see triggers and schedules.
Still stuck? We're happy to help.
Contact support