A thousand SaaS logins, one gateway, zero leaked API keys

What it does

OpenConnector sits between your AI agents and the SaaS apps they want to use. Users connect accounts once—via API key, OAuth2, or no-auth—and the gateway exposes a catalog of over 1,000 providers and 9,400 prebuilt Actions through an SDK, CLI, MCP, HTTP, or OpenAPI interface. The agents receive schemas, metadata, and execution results while the actual secrets stay behind the runtime boundary, whether it runs locally via Docker or Node.js, on Cloudflare Workers, or inside OOMOL’s hosted runtime.

The interesting bit

The project treats auth as infrastructure rather than an afterthought. It bundles a web console for credential management and run inspection, redacts logs, and supports policy controls like allow/block policies and runtime tokens so you can share connector access without sharing keys. The same provider IDs and Action contracts work across self-hosted and commercial deployments, which means you can prototype on Cloudflare and later retreat to a private runtime without rewriting agent code.

Key highlights

• Over 1,000 providers and 9,400 prebuilt Actions covering GitHub, Gmail, Notion, Slack, BigQuery, and others.
• Multiple agent interfaces: TypeScript SDK, local CLI relay, MCP endpoint, raw HTTP, and auto-generated OpenAPI.
• Credential isolation: secrets never reach the agent process; the gateway handles refresh, scopes, and encryption.
• Runs locally via Docker or Node.js with SQLite, on Cloudflare Workers with D1/R2, or through OOMOL’s managed runtime.
• Built-in web console for browsing providers, configuring connections, and inspecting redacted run logs.

Verdict

Worth a look if you’re building agent products that need durable, audited access to third-party tools. Skip it if your agents only call internal APIs and you already have a secrets manager you trust.