API Testing with AI Agents: Automate Endpoint Validation

Step-by-Step Setup

The following instructions configure HTTP Client Skill and OpenAPI Proxy as your foundation, then layer in Postman MCP and Newman CLI Skill for team and CI workflows. All four servers can coexist in a single config.

Step 1: Verify Node.js Installation

All API testing MCP servers are distributed as Node.js packages. Confirm you have Node 18 or later:

node --version  # should be v18 or higher

Step 2: Add Servers to Your MCP Config

Open your assistant's MCP configuration file. For Claude Code this is ~/.claude/settings.json; for Cursor it is .cursor/mcp.json:

{
  "mcpServers": {
    "http-client": {
      "command": "npx",
      "args": ["-y", "mcp-server-http-client"]
    },
    "openapi-proxy": {
      "command": "npx",
      "args": ["-y", "openapi-mcp-server"],
      "env": {
        "OPENAPI_SPEC_PATH": "./openapi.yaml"
      }
    },
    "postman": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postman"],
      "env": {
        "POSTMAN_API_KEY": "your_postman_api_key"
      }
    },
    "newman": {
      "command": "npx",
      "args": ["-y", "mcp-server-newman"]
    }
  }
}

Step 3: Point OpenAPI Proxy at Your Spec

Set OPENAPI_SPEC_PATH to a local file path or a public URL. The proxy supports OpenAPI 3.x YAML and JSON, as well as Swagger 2.0 documents. On startup it validates the spec and logs any parse errors before the agent can call any operations.

Step 4: Restart Your AI Assistant and Verify

Restart the assistant and confirm each server is connected. Useful test prompts:

• "Send a GET request to https://httpbin.org/get and show me the response headers" — verifies HTTP Client Skill
• "List all endpoints available in the loaded OpenAPI spec" — verifies OpenAPI Proxy
• "List my Postman workspaces" — verifies Postman MCP authentication
• "Run the newman health-check collection and summarize failures" — verifies Newman CLI Skill

Workflow: Define → Generate → Run → Report

The canonical AI-agent API testing workflow follows four phases. Each phase maps to one or more MCP skills in the stack described above.

Phase 1: Define

Start by giving the agent the API contract. If you have an OpenAPI spec, load it via OpenAPI Proxy and ask the agent to summarize all endpoints, their required parameters, and declared response schemas. If no spec exists, instruct the agent to construct one from API documentation or curl examples using the Swagger Parser MCP — this becomes your contract baseline.

Phase 2: Generate

Ask the agent to generate test cases: "For each POST endpoint in the spec, create three test cases — a happy path with valid inputs, a 400 case with a missing required field, and a 401 case with no Authorization header." The agent produces a Postman collection JSON that you can save locally or push directly to Postman Cloud via Postman MCP.

Phase 3: Run

Execute the generated collection. For interactive sessions, run it through Postman MCP so results appear inline. For CI pipelines, trigger Newman CLI Skill with the collection file and the target environment variables — Newman exits non-zero on any failing assertion, causing the pipeline to fail automatically.

Phase 4: Report

Ask the agent to parse Newman's JSON output and produce a human-readable summary: "Summarize the test run. List any failing assertions, the actual vs. expected values, and whether the failures look like regressions or environment issues." The agent can also format a Markdown table suitable for posting to a Slack channel or GitHub PR comment.

Comparison Table

Use this table to match each skill to your stage of the testing lifecycle and team workflow.

Frequently Asked Questions