About

Built for secure AI access to real systems

Swagger to MCP Gateway is a secure Model Context Protocol gateway for teams that want AI agents to use OpenAPI APIs and approved database scopes without exposing raw credentials, unrestricted endpoints, or direct database access.

The product is shaped by more than 10 years of enterprise architecture and production software delivery experience: integration design, identity boundaries, auditability, operational limits, and backend systems that must keep working after the demo.

Why we built this

AI agents need a safer way to use existing systems

Most companies already have useful APIs, databases, and internal workflows. The problem is that those systems were built for software clients and human operators, not for AI agents choosing tools from natural-language context.

Existing APIs should not be copied into prompts

OpenAPI specs are valuable, but raw endpoint lists, broad API keys, and vague operation comments are not enough for safe AI tool use. Teams need a smaller published surface with clear tool descriptions and server-side credentials.

Database access needs stronger boundaries

An agent should not receive a direct database connection just to answer business questions. Database access should be scoped by domain, table, column, executor mode, limits, and audit policy before any data is returned.

Governance should be repeatable

Every new AI client should not require a new security model. The Gateway exists so teams can publish, test, monitor, pause, and narrow MCP tools through one reviewed operational layer.

Mission

Make AI tool access governable

AI assistants are most useful when they can act on real systems. They are also risky when credentials, policy, and execution rules move into prompts or local client setup.

APIs become published MCP tools

Swagger/OpenAPI operations can be published as MCP tools with curated descriptions, selected operations, gateway-managed credentials, and usage visibility.

Databases stay behind scopes

PostgreSQL and SQL Server sources are exposed through approved database scopes, metadata-only schema tools, and validated executor tools.

Gateway remains the boundary

LLMs and MCP clients may propose actions, but Gateway-side validation, limits, redaction, and audit trails decide what actually executes.

Experience

Enterprise architecture, applied to MCP

The product is designed for B2B teams that need AI access to fit existing security, operations, and platform governance expectations.

Trust signals

10+ years
Enterprise architecture, integration design, identity boundaries, auditability, and production software delivery experience.
RTT Intelligence
Company context for applied AI, integration engineering, and secure software systems developed with an R&D discipline.
Gateway first
Security and validation remain in the backend Gateway, not inside model prompts.

Security by architecture

OpenAPI credentials and database connection strings are kept server-side. Published MCP tools are constrained by source ownership, publication state, limits, and execution policy.

Operational accountability

Tool calls need logs, usage metrics, latency data, failures, and quota visibility so teams can operate AI-driven traffic like normal production traffic.

Source-specific validation

OpenAPI calls, schema tools, QueryPlan executors, and SQL executors have different risks. The Gateway treats them as separate flows instead of one generic prompt problem.

Direct documentation

Public docs, LLM-readable files, and structured data describe the product in plain language so developers, search engines, and AI answer engines can understand it accurately.

Security commitments

What the Gateway will and will not expose

The security model starts from a practical assumption: prompts can be unclear, client configuration can drift, and agents can request the wrong thing. The Gateway keeps the enforcement layer in backend code where it can be tested and audited.

We do not treat the model as the security boundary. Tool descriptions help the model choose; Gateway validation decides what executes. That is why credentials, private route maps, database schemas, and customer data stay out of public LLM context, and why published tools can be hidden, tested, logged, and narrowed before wider rollout.

Schema tools are metadata-only

Database schema tools describe allowed tables, columns, relationships, keys, permissions, and operations. They do not return data row payloads.

Executor calls are validated

QueryPlan and SQL executor calls are checked against published scope, table, column, relationship, read-only, and owner boundaries before execution.

Control-plane specs stay private

The real management API is not published as a public root openapi.json. Public AI context is provided through llms.txt and the well-known product manifest.

Sensitive values stay out of public context

Credentials, connection strings, private route maps, database schemas, and customer data are not included in public LLM-readable files.

Evaluate the Gateway for your team

Review the setup docs, compare the security model, or contact us for B2B evaluation and partnership questions.