Every question
we get asked first.

Scraping services (e.g. Bright Data, ScrapingBee) send requests from datacenter IPs that are easily identified and blocked. McpFOX sends requests from your Chrome browser using your real TLS fingerprint, HTTP/2 stack, and personal cookies — achieving a 91% success rate against Cloudflare, PerimeterX, and Akamai, versus approximately 40–60% for typical scraping proxies. Additionally, scraping services charge $50–$300/month and route data through their servers. McpFOX is free for individuals and never transmits page content off your device.

AI assistants cannot directly access live web pages — their knowledge is frozen at a training cutoff date. Three specific problems follow: (1) AI returns outdated information when given a URL; (2) users spend $50–$300/month on scraping APIs to bridge this gap; (3) researchers manually copy-paste content between browser tabs and AI chat windows. McpFOX solves all three by making your Chrome browser the live web bridge, connected directly to your AI via MCP protocol.

Readability is Mozilla's open-source article extraction library — the same algorithm behind Firefox Reader Mode and Pocket. McpFOX applies it to every fetched page to remove navigation menus, advertisements, comment sections, related-post sidebars, and boilerplate. The result is clean article text only. This reduces AI token consumption by approximately 60% and improves the quality of summaries and answers by eliminating noise.

No. McpFOX is local-first by design. Fetched content lives only in two on-device layers: L1 (in-memory cache) and L2 (IndexedDB in your browser profile). There is no McpFOX relay server. The extension communicates from your Chrome instance directly to whichever MCP client is running locally (e.g. Claude Desktop at localhost:7431). You can disconnect from the internet entirely and still retrieve previously cached content from L2.

McpFOX requests three permissions: (1) activeTab — to read the URL of the active tab when triggered; (2) storage — to maintain the L2 IndexedDB cache; (3) nativeMessaging — to enable the local MCP server on port 7431. It does not request access to passwords, payment methods, full browsing history, location, microphone, or camera.

No. McpFOX reads page content at the network level — it does not have access to Chrome's password manager, autofill data, or payment methods. The activeTab permission scopes access to only the currently active tab's URL and rendered content, and only when an MCP tool explicitly requests a fetch for that URL.

Yes — this is one of its key capabilities. McpFOX inherits cookies from your Chrome session, so it can fetch paywalled articles, authenticated dashboards, internal tools, and any page your browser can reach without additional login steps. This is intentional design: your AI gains the same web access you have, nothing more.

The core fetch engine is MIT-licensed and publicly available on GitHub. You can read, build, and run the source directly. Chrome's extension review process requires the published CRX to match reviewed source. If you prefer maximum assurance, you can build from source and sideload as an unpacked extension in Chrome Developer Mode.

Four steps, approximately 2 minutes: (1) Install the McpFOX extension from the Chrome Web Store. (2) Open your Claude Desktop config file (claude_desktop_config.json). (3) Add: { "mcpServers": { "mcpfox": { "url": "http://localhost:7431" } } }. (4) Save and restart Claude Desktop. McpFOX will appear as a connected MCP server. No coding, no terminal required.

Open Cursor → Settings → MCP Servers → Add Server. Set the URL to http://localhost:7431 and save. McpFOX tools (fox_fetch, fox_extract, etc.) will appear in Cursor's tool panel. You can enable or disable individual tools from the same settings page.

No. The extension generates the configuration JSON for you at install time. For most AI clients, setup is four actions: copy, paste, save, restart. The Options panel lets you customise fetch templates, per-domain rules, and output format defaults through a GUI — no code required.

Chrome auto-updates extensions within 24 hours of a new Web Store release. To force an immediate update: open chrome://extensions/, enable Developer Mode, then click "Update". No config file changes or AI client restarts are needed for minor updates.

Right-click the McpFOX toolbar icon → Remove from Chrome. All cached data stored in IndexedDB is automatically purged on uninstall. Remove the "mcpfox" entry from your MCP client's config file to prevent connection errors on next launch.

MCP (Model Context Protocol) is an open standard published by Anthropic on 2025-03-26 for connecting AI language models to external tools and data. It uses JSON-RPC 2.0 as its message format and HTTP with Server-Sent Events (SSE) as its transport. MCP defines how AI clients discover available tools, call them with typed parameters, and receive structured results. McpFOX implements MCP version 2025-03-26 as a server at localhost:7431.

Six tools: fox_status (health check, version, port), fox_fetch (single URL → markdown/JSON/HTML/text/PDF), fox_fetch_batch (up to 200 URLs in parallel), fox_extract (structured data: title, author, date, word count, links, images), fox_cache_get (retrieve cached result by URL), fox_cache_list (list all cached items with metadata).

Yes. The local HTTP server at localhost:7431 accepts concurrent connections. Claude Desktop, Cursor, and Cline can all be connected simultaneously. Requests are handled by a priority queue with a default concurrency of 5 parallel fetch workers. There is no per-client limit.

No. Claude.ai (browser web interface) does not support MCP tool connections from browser extensions — it requires a local client application. McpFOX works with Claude Desktop (the downloadable app), not the claude.ai website. This is a constraint of Claude.ai's architecture, not of McpFOX.

McpFOX successfully returns parseable content for 91.4% of URLs tested against live anti-bot systems including Cloudflare Bot Management, PerimeterX, and Akamai Bot Manager. This is achieved through real Chrome TLS fingerprinting (JA3/JA4 matching Chrome 120+), native HTTP/2 frame ordering, rotating headers, and exponential backoff on retries. The remaining 8.6% of failures are pages requiring an active human-completed CAPTCHA, or pages that require login sessions not present in the browser.

Under 10ms for an L1 cache hit. 0.8–2.5 seconds for an uncached direct fetch, depending on the target server. JavaScript-rendered pages (Pro tier, SPA mode) add 1–3 seconds for browser rendering. Batch fetches of 50 URLs complete in approximately 8–15 seconds at the default concurrency of 5 workers.

No measurable impact during idle. The service worker is dormant when not serving requests. Active CPU usage during a fetch is one background thread for approximately 0.5–2.5 seconds. Idle RAM usage is under 2 MB. There is no impact on foreground tab performance.

In-flight requests return a CONNECTION_ERROR response to the AI client. Requests are not automatically retried. Reopening Chrome restores the server on port 7431 automatically — no manual restart required.

Yes. All six MCP tools, all five output formats, and baseline anti-bot protection are included at no cost. The only constraint is 200 fetches per day — roughly one every 7 minutes if used evenly throughout the day. No credit card required, no trial expiry, no crippled features.

Pro ($19/month) adds: unlimited daily fetches; rendered/streaming/multimodal fetch modes for JavaScript-heavy SPAs; 200-URL parallel batch; cloud history sync; per-domain custom rules; priority fetch lane; and email support with a 24-hour SLA.

Yes. All paid plans carry a 14-day unconditional refund policy. Email billing@mcpfox.space within 14 days of payment for a full refund — no reason required. Annual plans are refunded pro-rata for unused months.

Immediately on checkout confirmation. No manual provisioning required. Limits update in the extension automatically within seconds of payment processing.

Windows 10/11, macOS 12+, Ubuntu/Debian Linux (64-bit), and Chrome OS. McpFOX runs entirely within the Chrome extension context — there is no separate executable to install. Chrome version 110 or later is required.

Yes, in the Pro tier. The rendered fetch mode executes JavaScript fully before extracting content, covering React, Vue, Angular, Next.js applications, social media feeds, and other dynamic content. The free tier uses direct HTTP fetching — this works for server-rendered pages but does not execute JavaScript.

Claude Desktop, Cursor, Cline, Continue, ChatGPT Desktop, Windsurf, and any client implementing MCP 2025-03-26. Zed support is planned. Any tool speaking standard JSON-RPC 2.0 over HTTP can integrate without code changes.

The core fetch engine is MIT-licensed and on GitHub. Commercial components — cloud sync, team dashboards, enterprise integrations — are proprietary. This open-core model lets you audit the code handling your data, fork the engine for your own projects, and contribute to the community.

Yes. Third-party tools can connect to the local MCP server at localhost:7431 using standard MCP 2025-03-26 protocol. No special access or API key is required. The full protocol reference is in the MCP Technical Specification in the project docs.

Enterprise (starting at $399/month) includes: private on-premises deployment; SOC 2 compliance audit trails; contractual SLA guarantees; custom domain-rule libraries; a dedicated customer success manager; priority fetch lane; and onboarding training. Designed for regulated industries and larger teams. Contact sales@mcpfox.space for a custom quote.

The free and Pro tiers process no personal data on McpFOX servers — all data stays on your device. GDPR compliance therefore depends on your own data handling policies, not McpFOX's. Enterprise tier includes data processing agreements and audit logs for regulated environments.