Network Body Capture

Gasoline is an open-source MCP server that captures HTTP request and response bodies, letting AI coding assistants see the actual data flowing between your app and APIs. No more copy-pasting from DevTools.

The Problem

API debugging with AI assistants is painful. You see a 400 error, but without the response body your AI is guessing. You have to open DevTools, find the request, copy the response, paste it into chat — and repeat for every API issue.

With network body capture, your AI reads the payloads directly.

What Gets Captured

For every network request, Gasoline records:

Field	Description
Request body	POST/PUT/PATCH payloads (up to 8KB)
Response body	Server responses (up to 16KB)
Headers	Request and response headers (auth stripped)
Status code	200, 401, 500, etc.
Content type	JSON, HTML, text, etc.
Duration	Request timing in milliseconds

Filtering

Your AI can query exactly the requests it needs:

// Find failed auth requests
{
  "url": "/api/auth",
  "status_min": 400,
  "status_max": 499
}

// Find slow POST requests
{
  "method": "POST",
  "limit": 5
}

// Find requests to a specific endpoint
{
  "url": "users/profile"
}

MCP Tool: get_network_bodies

Parameter	Description
url	Filter by URL substring
method	Filter by HTTP method (GET, POST, etc.)
status_min	Minimum status code
status_max	Maximum status code
limit	Max entries to return (default: 20)

Example Response

{
  "url": "https://api.example.com/users",
  "method": "POST",
  "status": 422,
  "contentType": "application/json",
  "duration": 145,
  "requestBody": "{\"email\":\"invalid\",\"name\":\"\"}",
  "responseBody": "{\"errors\":{\"email\":\"invalid format\",\"name\":\"required\"}}"
}

Your AI immediately sees the validation errors without you lifting a finger.

Size Limits & Safety

Limit	Value
Request body cap	8KB (larger payloads truncated)
Response body cap	16KB
Buffer size	100 recent requests
Total memory budget	8MB
Auth headers	Always stripped

Privacy

• Auth headers stripped — tokens never appear in logs
• Localhost only — captured data stays on your machine
• Bounded buffers — old requests evicted, never unbounded growth

Use Cases

API Validation Errors

“My form submission is failing.”

Your AI sees the exact validation errors in the response body and can fix the request payload.

Authentication Issues

“I’m getting 401s.”

Your AI inspects the request to see what’s missing and the response for error details.

Data Format Mismatches

“The API returns data but the UI is empty.”

Your AI compares the response structure to what your code expects — spotting field name changes, missing properties, or type mismatches.

Debugging Third-Party APIs

“The Stripe webhook isn’t working.”

Your AI sees both what Stripe sent and how your server responded, identifying the disconnect.