Quilzo's MCP gateway puts your MCP servers and REST APIs behind one governed door. Any agent, from any vendor, reaches only the tools you pinned, with the access its role allows, and every call is recorded; the calls you choose wait for a person.
Why put a gateway in front of MCP
The Model Context Protocol made it easy to give agents tools. It also made it easy to give every agent every tool, with a shared credential and no record of who called what. A tool server can change a tool's description after you approved it, and a description is text a model will follow.
What the gateway does
- Pinned tools. Only the tools a person agreed to, and only while each tool's definition is the one they approved. A redefined tool is refused and reported.
- Access by role. Each tool server is offered to the roles you choose, with a daily count per caller.
- Credentials stay inside. Quilzo injects the server's credential; the calling agent never holds it.
- Approval call by call. Tools you mark wait for a person in Integrations, and a different administrator decides.
- Every call recorded. Who called which tool, on whose behalf, with a digest of the arguments, in the signed log.
- Private networks, declared. Internal MCP servers are reachable only at the address, port and path you declared.
REST APIs as tools
Import an OpenAPI 3 document and choose the operations to offer: each becomes a tool with the same pins, approvals and audit as an MCP tool. Calls carry only the declared parameters, over HTTPS, with Quilzo's credential. An MCP registry server.json becomes a server declaration the same way.
quilzo integrations import tracker.openapi.json --uses create_issue,list_issues
quilzo integrations gateway tracker --role author
# agents connect to https://your-quilzo/mcp/gateway/tracker
More than MCP
| Interface | What it is for |
|---|---|
/mcp | Quilzo's own operations for any agent or AI app, with OAuth 2.1 and a person's consent |
/mcp/gateway/NAME | Your tool servers, governed |
/v1 | OpenAI- and Anthropic-compatible model APIs, budgeted and masked, for agents built in any framework |
/a2a | Tasks from other agents over A2A 1.0, each under the receiving agent's declaration |
Agents built with LangChain, CrewAI, Google ADK, Microsoft Agent Framework or the OpenAI Agents SDK connect with a Quilzo token as their key, and each call is budgeted to its person and recorded.
