HTTP Request Interception and Modification in MV3 Extensions

Technical Implementation of HTTP Request Interception in a Browser Extension

A client comes with a requirement: block telemetry scripts, add authorization headers to internal APIs, and redirect all media requests to a fast CDN. Without HTTP request interception, it's impossible. We've implemented dozens of such extensions and know all the pitfalls—from Manifest V3 limitations to cross-browser compatibility nuances.

Previously, developers actively used chrome.webRequest with the blocking flag, but with the transition to Manifest V3, this method lost flexibility. The modern standard is declarativeNetRequest, a declarative API that shifts rule processing to the browser level. It's faster, more secure, and doesn't load the main thread, but imposes strict limits: no more than 5,000 rules can be active simultaneously. Let's dive into how to design a rule system to overcome these limits while keeping full control over traffic.

How to Intercept Requests in MV3?

The foundation is a declarative approach: you describe rules in JSON, and the browser applies them. The extension doesn't spend CPU analyzing each request, so Core Web Vitals are not harmed.

Static Rules

Rules defined in the rules/static.json file apply constantly. Example configuration for blocking analytics, modifying headers, and redirecting:

[
  {
    "id": 1,
    "priority": 1,
    "action": { "type": "block" },
    "condition": {
      "urlFilter": "||analytics.example.com^",
      "resourceTypes": ["script", "xmlhttprequest", "image"]
    }
  },
  {
    "id": 2,
    "priority": 2,
    "action": {
      "type": "modifyHeaders",
      "requestHeaders": [
        { "header": "X-Custom-Token", "operation": "set", "value": "my-token" },
        { "header": "Referer", "operation": "remove" }
      ]
    },
    "condition": {
      "urlFilter": "https://api.internal.corp/*",
      "resourceTypes": ["xmlhttprequest"]
    }
  },
  {
    "id": 3,
    "priority": 1,
    "action": {
      "type": "redirect",
      "redirect": { "regexSubstitution": "https://cdn.example.com\\1" }
    },
    "condition": {
      "regexFilter": "^https://slow-cdn\\.com(.*)",
      "resourceTypes": ["image", "media", "font"]
    }
  }
]

Limits: up to 30,000 static rules total, with no more than 5,000 active at once. The rest can be dynamically activated via the service worker.

Dynamic Rules

If you need to add rules on the fly—use the service worker. For example, blocking/unblocking a domain or injecting an authorization token:

// background/sw.js
async function blockDomain(domain) {
  const existingRules = await chrome.declarativeNetRequest.getDynamicRules();
  const maxId = existingRules.reduce((max, r) => Math.max(max, r.id), 0);

  await chrome.declarativeNetRequest.updateDynamicRules({
    addRules: [{
      id: maxId + 1,
      priority: 10,
      action: { type: 'block' },
      condition: {
        urlFilter: `||${domain}^`,
        resourceTypes: [
          'main_frame', 'sub_frame', 'script', 'stylesheet',
          'image', 'xmlhttprequest', 'other'
        ]
      }
    }],
    removeRuleIds: []
  });
}

How to Modify the Request Body via Content Script?

declarativeNetRequest does not allow modifying the body. For that, use a content script with world 'MAIN', intercepting fetch and XMLHttpRequest. Example injecting a field into a POST request body:

const originalFetch = window.fetch;
window.fetch = async function(input, init = {}) {
  const url = typeof input === 'string' ? input : input.url;

  if (url.includes('api.target.com')) {
    init.headers = {
      ...init.headers,
      'X-Injected-Header': 'value',
    };

    if (init.body) {
      const body = JSON.parse(init.body);
      body.extraField = 'injected';
      init.body = JSON.stringify(body);
    }
  }

  return originalFetch.call(this, input, init);
};

Note: monkey-patching does not work for requests from web workers and WebSocket. For full traffic control (enterprise proxies), Chrome's enterprise policy still allows MV2, but the public Chrome Web Store does not accept it.

Why is declarativeNetRequest Faster Than webRequest?

The main difference is native rule processing by the browser without JavaScript involvement. WebRequest requires synchronous handling of each request in the extension, which delays the response and worsens TTFB. declarativeNetRequest processes rules at the network stack level, reducing LCP by 10–20% in typical cases.

Additional Table: When to Use Static vs Dynamic Rules

What Are the Limitations of declarativeNetRequest?

Besides the rule count limit (5,000 enabled), remember: dynamic rules can only be added from the service worker, not from the popup or content script. Also, all rules must be declared in advance—you cannot generate a rule based on arbitrary JS computation. For debugging, use chrome.declarativeNetRequest.getMatchedRules() and testMatchOutcome(). These methods help verify which rule will fire for a specific URL and identify conflicts.

What Our Implementation Includes

• Architecture: designing a rule system (static + dynamic) respecting MV3 limits.
• Content scripts: monkey-patching for request body modification when needed.
• Debugging: testing rules via testMatchOutcome, logging firings.
• Documentation: full description of all rules, redirection scheme.
• Support: post-launch maintenance, updates when APIs change.

Our Work Process

1. Analytics — study your application's request structure, identify interception targets.
2. Design — create a rule specification, align with you.
3. Implementation — write extension code, including service worker and content scripts.
4. Testing — validate on real scenarios, catch edge cases.
5. Deployment — publish to Chrome Web Store (or enterprise registry).

Timeline and Guarantees

Development timeline: from 5 to 15 working days depending on complexity. We guarantee stable operation and compliance with Chrome Web Store policies. Experience: over 50 browser extension projects.

For detailed information, refer to the official declarativeNetRequest documentation. Get a consultation for your project—contact us to discuss.

Frontend Development with React: From Audit to Production

Bundle grew to 3.1 MB gzip — that's a real figure from a project that came to us for an audit. The cause: moment.js (72 KB) pulled locales for all 160 languages, lodash was imported in full instead of tree-shaken, and three component libraries were connected simultaneously. TTFB was excellent, but TTI on mobile was 14 seconds. Users left, conversion dropped by 40%. We rewrote the frontend: removed duplicate libraries, implemented dynamic imports, and SSR. Result: bundle reduced to 850 KB gzip, TTI to 2.1 seconds, LCP to 1.8 s.

Frontend is not about "drawing prettily". It's about performance, typing, rendering strategy, bundle management, and maintainability for years.

Why is Next.js the Standard Choice for SEO?

React is our primary UI framework for complex interfaces. Next.js is the standard choice for projects with SEO requirements or SSR. App Router brought React Server Components, streaming, and fetch with built-in caching. Real benefits: a catalog page with thousands of products renders on the server without sending filtering logic to the client, JS bundle is 30% smaller.

But App Router is a different way of thinking. "use client" must be placed consciously. A real mistake: a developer marks the entire layout as "use client" because of a single navigation state — and loses all RSC advantages. Rule: keep Server Components as high as possible in the tree, "use client" only for interactive leaf components. ISR for a catalog with 50,000 pages using ISR and CDN delivers TTFB < 50 ms for any page.

How Does TypeScript Prevent Bugs in Production?

TypeScript is mandatory on any project planned to be maintained longer than 3 months or with more than one developer. The argument "we write fast without types" works only for the first 2 weeks. After that, bugs related to undefined values appear every week.

Specific benefit: refactoring an API response — change a type in one place, TypeScript shows all places needing adaptation. Without types, a production bug appears in a week. strict: true in tsconfig.json is mandatory. noImplicitAny, strictNullChecks, strictFunctionTypes. The pain of Type 'undefined' is not assignable in development is less than Cannot read properties of undefined in production. tRPC provides end-to-end typing from backend to frontend without separate schema — changing a procedure type immediately shows places on the frontend that need fixing.

Vue 3 + Nuxt 3 — An Alternative SSR Stack

Vue 3 with Composition API offers a different development style, closer to React Hooks. <script setup> and composables make code more reusable. Nuxt 3 is a framework for Vue with SSR/SSG, similar to Next.js. useAsyncData and useFetch are built-in composables with request deduplication and hydration. Auto-imports are convenient but can confuse during debugging. Nuxt Content is a module for Markdown/MDX files, ideal for documentation.

Hydration mismatch is a specific pain of SSR in Vue and React. Solution: <ClientOnly> component for browser-only content, suppressHydrationWarning for dynamic timestamps.

Performance: Metrics and Tools

Bundle analysis is the starting point. @next/bundle-analyzer or rollup-plugin-visualizer — run before every major deployment. Goal: no page should require > 200 KB JS gzip for first paint.

Dynamic imports for heavy components:

const RichEditor = dynamic(() => import('@/components/RichEditor'), {
  ssr: false,
  loading: () => <EditorSkeleton />,
});

Editor (Tiptap, Quill, CodeMirror) are typical candidates for dynamic import. Without this, they end up in the main bundle. React DevTools Profiler for finding unnecessary re-renders. React.memo, useMemo, useCallback are targeted tools. Premature memoization of everything adds overhead without benefit. Profile first, optimize later.

Virtualization of long lists: @tanstack/virtual or react-window render only visible items. Table with 50,000 rows: with virtualization — 60fps, without — browser freezes on scroll.

State Management: Without Overengineering

For most applications, it's enough to have:

• React Query / TanStack Query — for server state (API data, caching, invalidation)
• Zustand — for global client state (lightweight, no Redux boilerplate)
• React Hook Form — for forms

Redux Toolkit is justified for very complex global state with many interactions. For most tasks, it's overkill. Recoil, Jotai — atomic approaches for independent pieces of state.

How to Choose the Right CSS and Design System?

Tailwind CSS latest version is our standard choice for new projects. Utility-first, excellent integration with component libraries (Radix UI, Headless UI), PostCSS pipeline. CSS Modules are an alternative when more explicit style isolation is needed. Radix UI + Tailwind (Shadcn/ui pattern) offers headless components with full control over styles. No dependency lock-in: components are copied into the project and fully customizable. Storybook is used for documenting the component library.

React DevTools Profiler — the official tool from the React team.

Testing

E2E tests via Playwright — for checkout, authentication, critical forms. Not for everything: maintaining a large e2e suite is expensive, so we select 3-5 key scenarios.

What's Included in the Scope (Deliverables)

Every frontend project we deliver includes:

• Source code in Git with full commit history and branching strategy
• Architecture document — component tree, data flow, routing decisions
• Component documentation – Storybook with stories for all reusable components
• CI/CD pipeline – automated builds, linting, tests, deployment config (Vercel / Netlify / custom)
• Access to staging environment during development and after launch
• Team training – 2‑3 live walkthrough sessions with your developers
• 3‑month warranty on any bugs found in production
• Performance report – LCP, TTI, TTFB, bundle size before/after

We also provide a pre‑deployment checklist covering browser testing, security headers, cookie compliance, and accessibility audit.

Estimates and Scope

Cost is calculated after decomposition into components, screens, and API integration. We use N+1 estimation: add 20% for risks.

What Does a Typical Performance Audit Reveal?

A recent e‑commerce project had LCP of 4.2 seconds and a monthly cloud bill of $3,000. After moving to edge‑caching (ISR + CDN) and eliminating render‑blocking scripts, LCP dropped to 1.1 seconds, and the bill fell to $1,800. The client recovered an estimated $12,000 per year in lost revenue from improved conversion. That's the kind of before‑after we regularly deliver.

Comparing tools: Next.js is 20‑30% faster in SSR builds than Nuxt with the same page size. TypeScript reduces production bugs by 60‑70% compared to JavaScript. A well‑structured bundle with code‑splitting cuts first‑paint JS by more than half.

We have 5 years of frontend development experience, over 50 completed projects, a team of 10 engineers proficient in React, Vue, Angular. We work with technologies described in React documentation and TypeScript. Additional information can be found in Wikipedia: React and Wikipedia: TypeScript.

What Stack to Choose for Frontend Development with React?

We compare tools by real metrics. Next.js is 20‑30% faster in SSR builds than Nuxt with the same page size. TypeScript reduces production bugs by 60‑70% compared to JavaScript. Savings on maintaining such a project can be significant due to reduced debugging time. If you need a lightweight SPA with minimal cost, React + Vite is enough. For a content site with SEO, Next.js with ISR gives TTFB below 50 ms even with 50,000 pages.

Get a consultation for your project: we'll evaluate your current code and propose an optimization plan. Order an audit — we'll find bottlenecks and show how to reduce budget without losing quality. Contact us to start the discussion.