MCP (Model Context Protocol)

Register MCP gateways and tools, control which agents can access which capabilities, and enforce policies on all MCP tool invocations in real time.

Overview

MCP (Model Context Protocol) enables AI agents to discover and use external tools and data sources. Without governance, MCP creates a large, ungoverned attack surface — agents can invoke shell commands, read files, query databases, and call external APIs.

Rivaro governs MCP at the gateway level: every tool invocation passes through enforcement, detection runs on inputs and outputs, and access is controlled by detection keys.

Gateways and Tools

Rivaro uses a two-tier MCP model:

Role	Description
Gateway	A governed MCP server entry point. Has its own detection key. Agents connect to the gateway — not directly to tools. The gateway controls which tools are accessible.
Tool	A specific capability bound to a gateway (e.g. file_read, database_query, web_search). Each tool has its own detection configuration and enabled detectors.

Registration Workflow

Step 1: Create a gateway

POST /api/admin/mcp/configurations

{
  "name": "Production MCP Gateway",
  "role": "gateway",
  "upstreamEndpoint": "https://your-mcp-server.internal/mcp",
  "enabledDetectors": [
    "PII Detection",
    "Credentials Detection",
    "Agent Tool Detection"
  ]
}

On creation, a detection key is generated and shown once. Copy it now — it cannot be retrieved again.

Step 2: Register tools

POST /api/admin/mcp/configurations

{
  "name": "file_read",
  "role": "tool",
  "boundGateway": "Production MCP Gateway",
  "enabledDetectors": [
    "PII Detection",
    "Credentials Detection"
  ]
}

Repeat for each tool. Tools that are not registered are not discoverable by agents connecting through the gateway.

Step 3: Configure your agent

Point your agent's MCP client at the Rivaro gateway with the detection key:

# Example: configuring an MCP client to use Rivaro as the gateway
mcp_client = MCPClient(
    gateway_url="https://your-org.rivaro.ai/mcp",
    headers={
        "X-Detection-Key": "detect_live_your_mcp_key_here"
    }
)

Configuration Fields

Field	Description
name	Gateway or tool name
role	gateway or tool
active	Whether this gateway/tool is enabled
upstreamEndpoint	The upstream MCP server URL (gateways only)
boundGateway	Which gateway this tool belongs to (tools only)
boundTools	List of tool names bound to this gateway (gateways only)
enabledDetectors	Risk categories to enforce (e.g. "PII Detection", "Agent Tool Detection")
detectorCount	Number of active detectors on this configuration

Access Control

Each gateway and tool has its own detection key. Only agents with the correct key can access a gateway or invoke a tool through it. Keys can be regenerated at any time — rotating a gateway key immediately revokes access for all agents using the old key.

# Regenerate a gateway's detection key
POST /api/admin/mcp/configurations/{configId}/regenerate-key

# Enable or disable a gateway/tool
POST /api/admin/mcp/configurations/{configId}/toggle

What Gets Detected

Every MCP tool invocation passes through the Rivaro detection engine. Tool calls generate AGENT_TOOL_* detection events that appear in the dashboard and in session history.

Rivaro also detects MCP-specific infrastructure risks via discovery:

Detection type	What it catches
INFRASTRUCTURE_MCP_PUBLIC_ENDPOINT	MCP server exposed to the internet without authentication
INFRASTRUCTURE_MCP_DANGEROUS_TOOLS	MCP server exposing shell execution, database writes, or other high-risk capabilities
INFRASTRUCTURE_SHADOW_AI_AGENT	MCP-connected agent running outside approved infrastructure

Viewing MCP Activity

MCP tool invocations are visible in session history as MCP_TOOL events. Each event records the tool name, arguments, and any detections triggered. Aggregate statistics — total invocations, active gateways, active tools — are available at:

GET /api/admin/mcp/stats

Managing MCP Configurations

Endpoint	Description
GET /api/admin/mcp/configurations	List all gateways and tools
POST /api/admin/mcp/configurations	Create a gateway or tool
PUT /api/admin/mcp/configurations/{configId}	Update configuration
DELETE /api/admin/mcp/configurations/{configId}	Delete (cascades to bound tools)
POST /api/admin/mcp/configurations/{configId}/toggle	Enable or disable
POST /api/admin/mcp/configurations/{configId}/regenerate-key	Rotate detection key
GET /api/admin/mcp/stats	Invocation statistics

Next steps

• Agent Tool Use detections — Full list of AGENT_TOOL_* detection types
• Sessions — How MCP_TOOL events appear in session history
• Discovery & Shadow AI — Find unregistered MCP servers in your environment
• Enforcement & Policies — Configure policies for tool invocation events