Enterprise MCP Gateway

What enterprise-mcp-gateway MCP server does

The enterprise-mcp-gateway MCP server places a policy and routing layer between MCP clients and internal services. It can expose operations described by OpenAPI 3.0 or Swagger specifications as MCP tools, avoiding separate glue code for each endpoint. The documented target backends include OpenAPI services, Java or Spring Boot applications, Go microservices, and relational databases.

Requests pass through authentication and role checks before execution. A token maps to a role, and that role determines which tools appear in tools/list and which calls are accepted by tools/call. This makes the gateway suitable for separating read-only, support, and administrative tool access.

How it works

The gateway is distributed as a Go binary with a JSON-RPC 2.0 MCP protocol engine. It supports standard stdio transport for local processes and HTTP Server-Sent Events for networked or distributed clients. The routing layer performs the MCP handshake and dispatches calls to registered backend operations.

Response data is inspected before it reaches the model. Built-in handling covers credit card numbers with Luhn validation, US Social Security numbers, email addresses, phone numbers, AWS keys, JWTs, GitHub personal access tokens, and configured JSON keys. Custom regular expressions can extend the sanitization rules.

Each tool execution produces structured JSON audit data. Records include caller identity, the selected tool, SHA-256 hashes of parameters, execution duration, status, and redaction counts. These records can be followed as a log stream and forwarded or processed for SIEM use.

Setup and configuration

Building requires Go 1.24 or newer. The repository provides separate build commands for the gateway and a mock backend server. A sample YAML file configures the server name, version, transport, host, and port, along with governance and sanitizer behavior.

Run the binary in its default stdio mode with a configuration path and token. For network access, select sse transport and provide a port. The repository also documents connecting Claude Desktop by adding the binary command, configuration path, and token to its MCP server configuration. Anthropic's MCP Inspector can verify either stdio or SSE operation.

Governance configuration defines token-to-role mappings, a default role, and wildcard-based allowed tool patterns such as list*, get*, or *. Sanitizer settings enable or disable masking, identify sensitive JSON keys, and add custom patterns.

Tools and capabilities

The enterprise-mcp-gateway MCP server provides infrastructure capabilities rather than a fixed business-tool catalog:

• Register MCP tools from OpenAPI and Swagger YAML or JSON documents.
• Map tool inputs to OpenAPI path, query, and body parameters.
• Filter tool discovery and enforce execution permissions by role.
• Authenticate calls with configured tokens.
• Mask PII, secrets, and selected JSON fields in streamed or JSON responses.
• Emit structured audit records with hashed parameters and latency data.

The actual business operations depend on the configured API specification and backend. The example configuration references operations such as customer listing, customer detail lookup, and support-ticket creation.

Limitations and notes

Those systems are identified as backend targets; their exposed operations still need to be represented through the supported configuration or API description.

No environment variables are specified in the example setup. Configuration is supplied through YAML and command-line options, including the token, config path, transport, and port. The README names Claude Desktop and Cursor as client examples and also mentions Antigravity and LangGraph, but it does not provide dedicated configuration instructions for the latter two.