What this solution does
The MCP Gateway provides controlled access for AI agents to your API. Using the Model Context Protocol (MCP), you can connect tools such as Claude, Codex, and Cursor, granting them access exclusively to the API methods you explicitly authorize. The gateway does not expose the site database to agents; their ability to read or modify data is determined entirely by your routes and their controllers.
The module operates only in conjunction with the API Platform. The MCP Gateway is an extension for the native.api module version 2.0.0 or newer; installation is impossible without this platform. The API Platform handles incoming agent requests, validates tokens, groups, and limits, while the Gateway manages the MCP protocol and the tool catalog.
The solution suits companies that want to give employees or clients access to site data and functions via AI without granting administrative rights. Developers can use it to publish existing API methods for agents without deploying a separate server. Integrators and studios can utilize this component as a reusable layer for client projects built on the API Platform.
To start the agent, install the module via the standard Marketplace wizard and add agent users to the MCP: Clients group.
Enable the gateway in the API platform settings and select the routes that will become tools.
Connect the agent to the gateway address using the user's access token.
Publishing custom tools requires a developer: data handling logic and permission checks for individual records are implemented in your project.
Capabilities for your team
Explicit publishing. Only routes marked for publishing become tools. Other API methods remain inaccessible to the agent.
Preserved access rules. The agent's request passes through the standard API platform pipeline: access token validation, user group checks, IP address restrictions, schedules, request limits, events, and caching.
Group-based separation. Different users see different sets of tools. An inaccessible tool cannot be invoked even by name.
Automatic argument schemas. The agent receives the description and schema for each tool from the route parameters, and the gateway validates arguments before execution.
Client hints. Tools can be marked as read-only, modifying, or external-system-facing—the agent client uses this information, for example, to request confirmation.
One-click disable. Revoke the token or remove the user from the "MCP: Clients" group.
The MCP gateway module by WEBCO has been tested with popular agents. Connection, catalog retrieval, and tool invocation have been verified in Codex and Claude. The documentation provides settings for Claude, Claude Desktop, Claude Code, Codex, VS Code, Cursor, and Gemini CLI, along with a ready-made instruction for automatic agent connection.
Version limitations:
The agent transmits the API platform access token in the request header.
Only tools are supported. Resources, prompts, tasks, and a dedicated SSE stream are not implemented.
The gateway has no internal tables, logs, or administrative pages: tokens, logs, and settings remain within the API platform.
Review the documentation before deployment. It includes:
Installation, access configuration, and user group management instructions.
Guides for connecting AI agents and agent-specific setup instructions.
Methods for publishing routes as tools.
Tool examples and a description of the JSON-RPC protocol.
General iT can install “MCP Gateway”, configure its settings and verify its operation in the existing project.