REST-to-MCP: Existing Business APIs as Agent Tools

Customer Benefit: Your existing business APIs become agent tools that only run after a decision, with consent and a confirmation for every write.

Overview

Modern enterprises run on REST APIs—order management, ERP, HR systems, and databases. Connecting these APIs directly to LLM agents is risky: language models frequently hallucinate parameter values, guess required identifiers, or execute destructive actions without confirmation.

Siliqun Rotor's REST-to-MCP gateway imports standard OpenAPI 3.0 and 3.1 specifications and serves them as MCP (Model Context Protocol) tools governed by a typed decision layer.

Comparison: Siliqun Rotor vs. AWS AgentCore Gateway

Capability	AWS AgentCore OpenAPI Gateway	Siliqun Rotor REST-to-MCP
Call Governance	Directly maps incoming requests to API calls via `operationId` without a preceding decision layer.	A typed decision (`decided`, `ask`, `refused`, `unresolved`) precedes every call. Missing arguments trigger an `ask` before the endpoint is touched.
Write Protection	Executes write operations directly based on model output.	Operations with `effect: "act"` require explicit confirmation (`act_needs_confirmation`) and execute with an idempotency key.
Consent & Scoping	Global API-level credentials.	Per-tool consent (`tool_scopes`). If an API declaration changes, execution is refused (`tool_declaration_changed`) until reviewed.
Deployment Model	Hosted exclusively inside AWS cloud infrastructure.	Runs where your applications run: hosted by us or self-hosted in your own environment on CPU without requiring an external LLM.

How It Works

Import OpenAPI: Upload or point Studio to an OpenAPI 3.0/3.1 JSON or YAML specification.
Declare Effects & Scopes:
effect: read (safe query) or act (state-modifying write).
scope: self (user's own data), org, or any.
Decision & Routing: When an incoming request is received:
If the request matches a read tool and all arguments are present, the engine returns decided.
If required parameters are missing, the engine returns ask.
If the tool performs a write action, it returns a proposal requiring user confirmation before execution.
SSRF-Guarded Execution: The gateway executes requests using strict IP allowlists and URL validation, preventing SSRF (Server-Side Request Forgery) attacks against internal metadata services or internal networks.

Try It in Studio

You can import an OpenAPI definition and inspect generated tool declarations in the Studio:

Studio Route: Navigate to Connect Tools.
API Endpoint: Served at POST /mcp/p/{project} with OAuth 2.1 or project API key.