Custom HTTP tools
Not every capability you want to give an agent lives behind an existing MCP server. Custom HTTP tools let you wrap a plain HTTP endpoint as an MCP tool, so an internal API becomes callable by any client connected to the Virtual MCP.
When to use them
Use a custom HTTP tool when:
- You have an internal or third-party REST API you want an agent to call.
- The API does not ship an MCP server of its own.
- You want tight control over the exact request the agent can make.
For a full MCP server you already run, register it as a connector instead.
Defining a tool
A custom HTTP tool describes the request the gateway will make on the agent’s behalf:
- Name and description: what the agent sees in
tools/list. A clear description is the most important factor in whether the model uses the tool correctly. - Method and URL: the HTTP method and endpoint to call. URLs support parameter placeholders.
- Input schema: the parameters the agent must provide, mapped into the path, query string, headers, or body.
- Authentication: static headers (for example an API key) applied to every call.
When the agent invokes the tool, the gateway builds the request from the schema-validated arguments, calls the endpoint, and returns the response body as the tool result.
Security notes
- Scope each tool to the narrowest endpoint that does the job. Prefer a specific path over a broad, templated URL.
- Keep credentials in the tool’s header configuration rather than embedding them in the URL.
- Custom HTTP tools run with whatever access their configured credentials grant. Treat them like any other credential you hand to an agent.
Next steps
- MCP endpoints shows how tools are aggregated and namespaced.
- Webhook triggers covers driving agents from inbound events.