Fast WebFetch MCP Server for Claude Code

A high-performance MCP server for web fetching in Claude Code using Firecrawl backend with automatic fallback.

Features

Tool	Description
fast_fetch	Fetch URL, extract content, return markdown
fast_fetch_raw	Return raw HTML without processing
fast_fetch_json	Fetch and parse JSON endpoints
fast_fetch_multiple	Fetch multiple URLs in parallel

How It Works

1. Primary: Uses Firecrawl API for JS rendering and comment extraction
2. Fallback: If Firecrawl unavailable, uses direct fetch + Mozilla Readability
3. AI Summary: Optional summarization via local LLM API

URL -> Try Firecrawl -> Success? -> AI Summary (optional) -> Result
              |
              v (failed/unavailable)
       Direct Fetch + Readability -> AI Summary (optional) -> Result

fast_fetch Parameters

Parameter	Type	Default	Description
url	string	required	The URL to fetch
prompt	string	-	AI summarization prompt
max_length	number	100000	Maximum content length (chars)
include_links	boolean	true	Include hyperlinks in output
no_summarize	boolean	false	Skip AI summarization even with prompt

Quick Install

Using Claude CLI (Recommended)

# Clone the repo
git clone https://github.com/nikketryhard/fast-webfetch-mcp.git ~/fast-webfetch-mcp

# Install dependencies
cd ~/fast-webfetch-mcp && bun install

# Add MCP server with environment variables
claude mcp add fast-webfetch \
  -e FIRECRAWL_API_URL=http://localhost:3002 \
  -e FAST_WEBFETCH_API_URL=http://127.0.0.1:8045/v1 \
  -e FAST_WEBFETCH_MODEL=gemini-3-flash \
  -e FAST_WEBFETCH_API_KEY=your_api_key_here \
  -e FAST_WEBFETCH_USE_RUST=1 \
  -e FAST_WEBFETCH_RUST_BIN=/home/YOUR_USERNAME/fast-webfetch-mcp/rust/target/release/fast-webfetch-mcp-rs \
  -- bun run ~/fast-webfetch-mcp/src/index.ts

# Or minimal (uses defaults)
claude mcp add fast-webfetch -- bun run ~/fast-webfetch-mcp/src/index.ts

Manual Installation

# Clone the repo
git clone https://github.com/nikketryhard/fast-webfetch-mcp.git ~/fast-webfetch-mcp

# Install dependencies
cd ~/fast-webfetch-mcp && bun install

Then add to ~/.claude.json:

{
  "mcpServers": {
    "fast-webfetch": {
      "command": "bun",
      "args": ["run", "/home/YOUR_USERNAME/fast-webfetch-mcp/src/index.ts"],
      "env": {
        "FIRECRAWL_API_URL": "http://localhost:3002",
        "FAST_WEBFETCH_API_URL": "http://127.0.0.1:8045/v1",
        "FAST_WEBFETCH_MODEL": "gemini-3-flash",
        "FAST_WEBFETCH_API_KEY": "your_api_key_here",
        "FAST_WEBFETCH_USE_RUST": "1",
        "FAST_WEBFETCH_RUST_BIN": "/home/YOUR_USERNAME/fast-webfetch-mcp/rust/target/release/fast-webfetch-mcp-rs"
      }
    }
  }
}

Environment Variables

Variable	Default	Description
FIRECRAWL_API_URL	http://localhost:3002	Firecrawl API endpoint
FAST_WEBFETCH_API_URL	http://127.0.0.1:8045/v1	OpenAI-compatible API for summaries
FAST_WEBFETCH_MODEL	gemini-3-flash	Model name for summarization
FAST_WEBFETCH_API_KEY	(empty)	API key for summarization endpoint
FAST_WEBFETCH_USE_RUST	0	Enable Rust backend worker path
FAST_WEBFETCH_RUST_BIN	(empty)	Absolute path to Rust binary

Firecrawl Setup (Optional but Recommended)

For best results (JS rendering, comments extraction), run Firecrawl locally:

# Using Docker
docker run -p 3002:3002 mendableai/firecrawl

# Or use the official Firecrawl MCP
claude mcp add firecrawl-mcp -e FIRECRAWL_API_URL=http://localhost:3002 -- npx -y firecrawl-mcp

Without Firecrawl, the server falls back to direct fetch + Readability (no JS rendering).

Comparison with Native WebFetch

Feature	Native WebFetch	Fast WebFetch
Reddit/Twitter	Blocked (403)	Works
User agent	Claude-User (blocked)	Real browser UA
JS rendering	No	Yes (with Firecrawl)
Comments	No	Yes (with Firecrawl)
Content	Haiku-summarized	Full markdown
AI summary	Haiku 3.5	Configurable model
Max length	100KB	Configurable
Raw HTML	No	Yes
JSON fetch	No	Yes
Parallel fetch	No	Yes
Fallback	None	Direct fetch

Requirements

• Bun runtime
• Claude Code 2.0+
• (Optional) Firecrawl for JS rendering
• (Optional) OpenAI-compatible API for summarization

License

MIT

Contributing

PRs welcome! Please open an issue first to discuss changes.