Why a gateway
An MCP gateway fronts many MCP servers behind one authenticated, governed endpoint. Instead of configuring every server in every AI client, you attach them once to BusinessMCP and reach them all through your single /api/mcp URL — with unified auth, per-key access policies and output redaction.
Connect a server
In Connections → Custom, add an MCP server: paste its URL (streamable HTTP) and, if it needs auth, a bearer token. The token is stored server-side in your vault. That's it — the server is now part of your gateway.
In the Assistant
The connected server's tools are available to the Assistant automatically. On Anthropic models they attach natively; on other models (Gemini, GPT, Groq, the free tier) they are aggregated as discrete namespaced tools (ext_<server>_<tool>) so every model can use them.
Through your endpoint
External agents hitting your endpoint reach attached servers through two gateway tools:
- `list_connected_mcp_tools` — discover the tools across your connected servers.
- `call_connected_mcp_tool` — invoke one by its namespaced name with an arguments object.
Output is scrubbed by the calling key's access policy, the same as any other tool.
Governance & reliability
Every upstream call is time-boxed and isolated: a dead server is skipped, never fatal. Access is gated by policy, and results are redacted for revenue, PII and enrichment when the key isn't permitted them. You get the reach of many servers with the control of one endpoint.
Frequently asked questions
Does a slow or offline external server break my runs?
No. Every gateway call is time-boxed and best-effort. If an attached MCP server is unreachable, it is simply skipped — it never blocks your platform tools, connectors or the rest of the run.
Keep going
Turn your company into one AI-ready data platform on a single hosted MCP endpoint.