Skip to main content
function safetyCheck(
  handler: (params: any) => Promise<any> | any,
  options: SafetyCheckOptions,
): (params: any) => Promise<any>
A higher-order function that wraps an MCP tool handler so that one of its input parameters is validated against Sedata’s safety API before the handler runs. Flagged input short-circuits with a structured “blocked” response.

​
Parameters

​
handler
(params: any) => Promise<any> | any
required
Your original tool handler, exactly as you’d pass to server.registerTool.
​
options
SafetyCheckOptions
required
Configuration for the wrapper.

​
SafetyCheckOptions

​
parameterName
string
required
Name of the parameter on the handler input to validate. Only string values are checked — non-string values are ignored.
​
output_screen
boolean
When true, prints a 🚨 Safety Check Alert: ... line to stdout when content is flagged. Useful for local debugging. Defaults to false.
​
endpoint
string
Reserved for a future override of the safety endpoint. Currently the wrapper always calls https://api.sedata-ai.tech/security/safety-check.

​
Returns

A new handler with the same signature. Internally:
  • Reads params[options.parameterName].
  • If it’s a string, calls checkSafety(value) and writes the result to params._safetyCheck so the instrumentation layer can attach mcp.safety_check.* attributes.
  • If flagged, returns a structured blocked response without calling the original handler.
  • Otherwise calls the original handler with params.

​
Blocked response

{
  "content": [
    {
      "type": "text",
      "text": "{ \"summary\": \"🚫 CONTENT BLOCKED: ...\", \"blocked\": true, \"reason\": \"Content flagged by safety check\", \"timestamp\": \"...\" }"
    }
  ],
  "structuredContent": {
    "summary": "🚫 CONTENT BLOCKED: ..."
  }
}
The structuredContent.summary field is what most tools surface to the caller. If your tool’s outputSchema uses a different key name, you’ll need to fork or adapt the wrapper.

​
Example

import { z } from 'zod'
import { safetyCheck } from '@sedata-ai/mcp'

server.registerTool(
  'text-summarizer',
  {
    title: 'Text Summarizer',
    description: 'Summarize text content',
    inputSchema: { text: z.string() },
    outputSchema: { summary: z.string() },
  },
  safetyCheck(
    async ({ text }) => {
      const summary = text.substring(0, 100) + '...'
      return {
        content: [{ type: 'text', text: JSON.stringify({ summary }) }],
        structuredContent: { summary },
      }
    },
    { parameterName: 'text', output_screen: true },
  ),
)

​
Authentication

The wrapper uses the API key set during instrumentServer(...). Both exporterAuth.apiKey and exporterAuth.token are forwarded as the x-api-key header. To override: 
import { setTelemetryApiKey } from '@sedata-ai/mcp'
setTelemetryApiKey('different-key')

​
Failure modes

FailureBehavior
Network errorLogged. Handler runs. _safetyCheck.success = false.
Non-200 from safety APILogged. Handler runs. _safetyCheck.success = false.
parameterName not a string or missingHandler runs without check.
The wrapper fails open — telemetry should never break your tool.

​
Span attributes added

When the wrapper runs, the active tool span receives:
AttributeType
mcp.safety_check.contentstring
mcp.safety_check.flaggedboolean
mcp.safety_check.latency_msnumber
mcp.safety_check.successboolean

​
See also

checkSafety

The lower-level function that talks to the safety API.

Safety checks concept

What it means and when to use it.