API & CLI Integration

API & CLI Integration

📌 TL;DR: Understand how to trigger prompts and flows programmatically, including REST endpoints and CLI tools to embed MCP into CI/CD workflows.

RESTful & GraphQL Endpoints

Why it matters: Direct HTTP APIs will enable developers to trigger MCP flows from external systems, dashboards, or automation pipelines. Example Endpoint (REST): POST /mcp/execute { “task”: “threat_summary”, “input”: “CVE-2025-20345”, “model”: “claude-4-sonnet” } Expected Features:

• Auth via API keys or OAuth2
• JSON schema validation
• Streaming support (for long responses)
• Role-based access control per route

CLI Interface for DevOps & Analysts

Use Case: Security engineers want to run prompts or trigger workflows from their terminal or CI/CD tools. Example Usage: mcp exec summarize-threat –input “Stealer logs from 45.66.23.1” Supported Flags:

• –model (e.g., claude, llama3)
• –context-file (e.g., indicators.json)
• –dry-run (validate without execution)
• –output-format=json|markdown|pdf

Use in CI Pipelines: – name: MCP Lint & Risk Check run: | mcp exec compliance-check –input-file ./policy.json

Webhook & Async Execution

Problem: LLM-based tasks can be long-running. API should support async responses and callbacks. Planned Structure:

• Submit → get job_id
• Poll: GET /mcp/status/:job_id
• Or provide a callback_url and MCP notifies upon completion

Planned SDKs (Python, Go, JS)

Roadmap items:

• Python SDK for fast dev
• TypeScript SDK for frontend integration
• Go SDK for SOC/EDR tooling

Example Python usage: from mcp_sdk import MCPClient client = MCPClient(api_key=”s3cr3t”) response = client.execute(“intel_summary”, input=”Axiom APT in Eastern Europe”) print(response.summary)

Security & Governance for APIs

Key Features Expected:

• IP allowlisting
• Prompt integrity checks (hash match)
• Throttling & quota per user or API key
• Audit log for every CLI/API call

Bonus: Example Swagger/OpenAPI v3.0 for MCP Server

openapi: 3.0.3 info: title: SOCRadar MCP Server API description: API for triggering prompt-based tasks via the MCP Server. version: 1.0.0 paths: /mcp/execute: post: summary: Execute a prompt-based task description: Triggers an MCP Server workflow with specified model and context. operationId: executeTask requestBody: required: true content: application/json: schema: $ref: ‘#/components/schemas/ExecuteRequest’ responses: ‘200’: description: Successful task execution content: application/json: schema: $ref: ‘#/components/schemas/ExecuteResponse’ ‘400’: description: Invalid input ‘401’: description: Unauthorized ‘500’: description: Server error components: schemas: ExecuteRequest: type: object required: – task – input properties: task: type: string example: threat_summary description: The name of the MCP task or flow to run input: type: string example: CVE-2025-20345 description: Raw input prompt or data model: type: string example: claude-3-sonnet description: Optional model override context_file: type: string example: indicators.json description: Optional context file to preload dry_run: type: boolean example: false description: If true, validates the flow without executing ExecuteResponse: type: object properties: job_id: type: string example: job_abc123xyz status: type: string example: queued output: type: string example: “The CVE is associated with ransomware activity in Europe…”