Introduction
What PopMCP is, the problem it solves, and how the pieces fit together.
PopMCP is a hosted gateway that exposes the SaaS tools your business uses as Model Context Protocol tools, so an AI client can read and act on them directly.
The problem it solves
MCP clients can talk to remote servers. Wiring up a real business that way means running one server per tool, storing an API key for each in plaintext client config, and repeating that for everyone on the team. Rotating a key means editing every laptop. Removing someone's access means hoping they delete a file.
PopMCP replaces that with one authenticated URL.
How it works
You connect a tool with your own credentials
Paste a Klaviyo private key, or authorize a Shopify app you own. PopMCP encrypts the credential with AES-256-GCM and verifies it against the provider before saving.
PopMCP provisions an MCP surface for it
Each verified connection gets a curated toolset for that provider, plus discovery tools for reaching the rest of the provider's API.
Your AI client authenticates once
The client registers itself, runs an OAuth 2.1 flow with PKCE, and you choose which connectors to grant it. It then calls a single endpoint and sees every tool it was granted.
What makes it different
One URL, every tool. POST /mcp serves every customer. The gateway resolves who you
are on each request and aggregates exactly the connections you were granted.
Credentials never reach the client. Your AI client holds an OAuth token scoped to PopMCP. It never sees the Klaviyo key or the Shopify token underneath.
Access is re-resolved per request. Grants are not trusted from the token. Every call re-checks that the connection is still active, still verified, still yours, and still allowed. Revoking access takes effect on the next request.
Read-only is enforced server-side. A connector set to read-only filters its own tool list and blocks writes through the escape hatch, so a misbehaving agent cannot write.