MCP Server Overview

The Model Context Protocol (MCP) is a standard protocol for communication between AI assistants and external tools. Reqs can run as an MCP server, allowing AI assistants like Claude to send HTTP requests through it.

What is MCP Mode?

MCP mode transforms Reqs into a server that listens for requests from AI assistants. Instead of reading URLs from stdin, Reqs waits for the AI assistant to send requests via the MCP protocol.

Starting the MCP Server

To start Reqs in MCP mode, use the --mcp flag:

reqs --mcp

The server will start and wait for connections from MCP clients.

Server Options

You can configure the MCP server with various options:

Custom Timeout

reqs --mcp --timeout 30

Custom Headers

reqs --mcp --headers "Authorization: Bearer token123"

Concurrency Control

reqs --mcp --concurrency 20

HTTP/2 Support

reqs --mcp --http2

Combining Options

reqs --mcp \
  --timeout 30 \
  --concurrency 20 \
  --headers "User-Agent: AI-Assistant/1.0" \
  --http2

Available Tools

When running as an MCP server, Reqs provides the following tool:

send_requests

The send_requests tool accepts a list of HTTP requests and returns their results.

Input Parameters:

requests (required): Array of URLs or request strings in format METHOD URL BODY
filter_status (optional): Filter results by HTTP status codes (e.g., [200, 404])
filter_string (optional): Filter results containing specific text in response body
filter_regex (optional): Filter results matching regex pattern in response body
follow_redirect (optional): Whether to follow HTTP redirects. Defaults to true
http2 (optional): Use HTTP/2 for requests. Defaults to false (HTTP/1.1)
headers (optional): Custom headers to add to the request (e.g., ["User-Agent: my-app", "Authorization: Bearer token"])
include_req (optional): Include raw HTTP request details in output
include_res (optional): Include response body in output

Example Input:

{
  "requests": [
    "https://www.hahwul.com",
    "https://www.hahwul.com/about/",
    "https://github.com",
    "POST https://www.hahwul.com a=d"
  ],
  "filter_status": [200, 201],
  "filter_string": "example text",
  "filter_regex": "pattern.*match",
  "follow_redirect": true,
  "http2": false,
  "headers": ["User-Agent: my-app", "Authorization: Bearer token"],
  "include_req": true,
  "include_res": true
}

Output Format:

Basic output includes:

{"content_length":131,"method":"POST","response_time_ms":42,"status_code":405,"url":"https://www.hahwul.com"}
{"content_length":32498,"method":"GET","response_time_ms":43,"status_code":200,"url":"https://www.hahwul.com"}
{"content_length":30063,"method":"GET","response_time_ms":44,"status_code":200,"url":"https://www.hahwul.com/about/"}
{"content_length":0,"method":"GET","response_time_ms":49,"status_code":200,"url":"https://github.com"}

With include_req and include_res, the output includes additional fields:

{"content_length":149,"ip_address":"127.0.0.1","method":"GET","raw_request":"GET /path HTTP/1.1\nHost: example.com\n","response_body":"<html>...</html>","response_time_ms":42,"status_code":200,"url":"https://example.com"}

Use Cases

MCP mode is ideal for:

AI-Powered Security Testing: Let AI assistants discover and test endpoints
Automated Reconnaissance: AI can analyze responses and make intelligent follow-up requests
Interactive Analysis: Chat with an AI while it tests web applications in real-time
Dynamic Testing Workflows: AI can adapt testing strategies based on responses

Benefits

Natural Language Interface: Describe what you want to test in plain language
Context-Aware: AI assistants can analyze responses and make intelligent decisions
No Scripting Required: Complex testing workflows without writing code
Interactive: Real-time feedback and conversation during testing