What is an MCP server?
MCP (Model Context Protocol) is the open standard AI agents use to call external tools. An MCP server hosts those tools. AllMCP is a remote MCP server that aggregates every integration behind one URL — your agent connects once instead of running a separate server per product.
Which AI clients work with AllMCP?
Any MCP client. Claude (desktop, web, mobile), Claude Code, ChatGPT, Codex and Cursor are documented step by step in Setup.
Do I need an API key?
No. Add the connector and sign in with Google or GitHub when your browser opens. You only need an API key if you're building software and integrating AllMCP into your own agents — see compatible agentic frameworks ↗.
What counts against the 20,000 free calls?
Only real provider tool calls. System and discovery calls — listing providers, browsing tools, connecting accounts — are always free.
How are my SaaS credentials stored?
Encrypted at rest and never written to logs. Disconnecting a provider revokes them immediately; deleting your account erases them after the 7-day grace period. Details on the Security page.
How is this different from running MCP servers myself?
One hosted MCP server replaces a server per product: nothing to install, update or keep alive. We maintain the connectors; you get one endpoint, one sign-in and one quota. The longer answer — token efficiency, built-in skills, multi-tenancy — is in why we made AllMCP.
Can I build AllMCP into my own product?
Yes. AllMCP is multi-tenant by design: teams building AI agents connect their whole user base through the same endpoint — pass your organization and user IDs on the connect URL and each user's provider credentials stay isolated. You keep one API key; every one of your users gets their own connections.