StripFeed MCP Server

MCP server for StripFeed - convert any URL to clean, token-efficient Markdown.

Works with Claude Code, Cursor, Windsurf, and any MCP-compatible client.

Setup

1. Get your API key

Sign up at stripfeed.dev and create an API key. Keys start with sf_live_.

2. Install

Claude Code

claude mcp add stripfeed -- npx -y @stripfeed/mcp-server

Then set your API key:

export STRIPFEED_API_KEY=sf_live_your_key

Or add it to your shell profile (~/.zshrc, ~/.bashrc).

Cursor / VS Code

Add to your MCP settings (.cursor/mcp.json or VS Code MCP config):

{
  "mcpServers": {
    "stripfeed": {
      "command": "npx",
      "args": ["-y", "@stripfeed/mcp-server"],
      "env": {
        "STRIPFEED_API_KEY": "sf_live_your_key"
      }
    }
  }
}

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "stripfeed": {
      "command": "npx",
      "args": ["-y", "@stripfeed/mcp-server"],
      "env": {
        "STRIPFEED_API_KEY": "sf_live_your_key"
      }
    }
  }
}

Tools

fetch_url

Convert a single URL to clean Markdown. Strips ads, navigation, scripts, and boilerplate. Returns clean content ready for LLM consumption.

Parameters:

Example: Basic fetch

Prompt: "Fetch https://react.dev/learn and summarize the key concepts"

The tool calls fetch_url with url: "https://react.dev/learn" and returns:

Title: Quick Start - React | Tokens: 2,450 (saved 91.2% from 27,840) | Cache: MISS | Fetch: 342ms

---

# Quick Start

Welcome to the React documentation! This page will give you an introduction to 80% of the React concepts...

Example: Extract with CSS selector

Prompt: "Get the pricing table from https://stripe.com/pricing using selector '.pricing-table'"

The tool calls fetch_url with url: "https://stripe.com/pricing" and selector: ".pricing-table" and returns only the content matching that CSS selector.

Example: Plain text output

Prompt: "Fetch https://example.com as plain text"

The tool calls fetch_url with url: "https://example.com" and format: "text". Returns plain text with all Markdown formatting stripped (no headings, bold, links, etc.).

Example: Truncate to token budget

Prompt: "Fetch https://en.wikipedia.org/wiki/Rust_(programming_language) but limit to 500 tokens"

The tool calls fetch_url with url: "https://en.wikipedia.org/wiki/Rust_(programming_language)" and max_tokens: 500. The output is truncated at a paragraph or sentence boundary to fit within 500 tokens. The response metadata shows Truncated: yes.

Example: Custom cache TTL

Prompt: "Fetch https://news.ycombinator.com with a 2-hour cache"

The tool calls fetch_url with url: "https://news.ycombinator.com" and ttl: 7200. The content is cached for 7200 seconds (2 hours) instead of the default 3600 (1 hour). Subsequent requests within that window return instantly from cache.

Example: Bypass cache

Prompt: "Fetch https://news.ycombinator.com with fresh content, don't use cache"

The tool calls fetch_url with url: "https://news.ycombinator.com" and cache: false. Forces a fresh fetch even if cached content exists.

batch_fetch

Fetch multiple URLs in parallel and convert them all to clean Markdown. Process up to 10 URLs in a single call. Requires Pro plan.

Parameters:

Example: Compare multiple pages

Prompt: "Fetch the React, Vue, and Svelte docs and compare their approaches to state management"

The tool calls batch_fetch with urls: ["https://react.dev/learn", "https://vuejs.org/guide/introduction.html", "https://svelte.dev/docs/introduction"] and returns all three pages as Markdown sections:

Fetched 3/3 URLs successfully.

---

## Quick Start - React

Source: https://react.dev/learn | 2,450 tokens (saved 91.2% from 27,840)

# Quick Start
...

---

## Introduction - Vue.js

Source: https://vuejs.org/guide/introduction.html | 1,890 tokens (saved 88.5% from 16,400)

# Introduction
...

---

## Introduction / Svelte

Source: https://svelte.dev/docs/introduction | 1,120 tokens (saved 93.1% from 16,200)

# Introduction
...

Handling failures: If a URL fails (unreachable, timeout, invalid), it gets a per-item error and the other URLs still succeed:

Fetched 2/3 URLs successfully.

---

## Quick Start - React

Source: https://react.dev/learn | 2,450 tokens (saved 91.2% from 27,840)

# Quick Start
...

---

## https://invalid-url.example

Error: Target URL unreachable

---

## Introduction - Vue.js

Source: https://vuejs.org/guide/introduction.html | 1,890 tokens (saved 88.5% from 16,400)

# Introduction
...

Note on selectors: The batch_fetch tool sends plain URL strings. To use CSS selectors on individual URLs in a batch, use the REST API directly:

curl -X POST "https://www.stripfeed.dev/api/v1/batch" \
  -H "Authorization: Bearer sf_live_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "urls": [
      "https://example.com",
      {"url": "https://docs.anthropic.com", "selector": "article"},
      {"url": "https://stripe.com/pricing", "selector": ".pricing-table"}
    ]
  }'

Each URL item in the array can be a plain string or an object with url and optional selector.

Strategy: Efficient batch_fetch for Product Pages with Minimal Token Usage

To efficiently extract only relevant information from a list of product pages while minimizing token usage for an AI model, combine batch_fetch with these strategies:

Step 1: Use batch_fetch to fetch all pages at once

Call batch_fetch with all product URLs (up to 10 per call):

batch_fetch with urls: [
  "https://example.com/product/a",
  "https://example.com/product/b",
  "https://example.com/product/c",
  "https://example.com/product/d",
  "https://example.com/product/e"
]

This fetches all 5 pages in parallel in a single API call, rather than 5 sequential fetch_url calls.

Step 2: If pages are too large, re-fetch with selectors via the REST API

If batch_fetch returns pages with high token counts (e.g., 10K+ tokens each), use the REST API batch endpoint with per-URL CSS selectors to extract only the product section:

curl -X POST "https://www.stripfeed.dev/api/v1/batch" \
  -H "Authorization: Bearer sf_live_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "urls": [
      {"url": "https://example.com/product/a", "selector": ".product-details"},
      {"url": "https://example.com/product/b", "selector": ".product-details"},
      {"url": "https://example.com/product/c", "selector": ".product-details"}
    ],
    "model": "claude-sonnet-4-6"
  }'

Step 3: Use max_tokens on individual pages if still too large

For pages that are still large after selector extraction, use fetch_url with max_tokens to cap the output:

fetch_url with url: "https://example.com/product/a", selector: ".product-details", max_tokens: 1000

This truncates at a paragraph or sentence boundary to stay within 1,000 tokens.

Token savings summary:

• Raw HTML page: ~15,000 tokens

• After StripFeed Markdown conversion: ~1,500 tokens (90% saved)

• With CSS selector: ~500 tokens (97% saved)

• With selector + max_tokens: ~200 tokens (99% saved)

Monitoring usage: Call check_usage before and after batch operations to track how many API requests you have remaining on your plan.

check_usage

Check your current monthly API usage and plan limits. Takes no parameters.

Example response:

Plan: pro
Usage: 1,250 / 100,000
Remaining: 98,750
Resets: 2026-04-01T00:00:00.000Z

For free plan users:

Plan: free
Usage: 42 / 200
Remaining: 158
Resets: 2026-04-01T00:00:00.000Z

For enterprise plan users (unlimited):

Plan: enterprise
Usage: 54,200 (unlimited)
Resets: 2026-04-01T00:00:00.000Z

Handling fetch_url Failures and Empty Results

When fetch_url fails to process a URL or returns empty content, the MCP server throws an error with a specific status code and message. Here is how an MCP client should interpret and handle each failure mode:

Network and target failures:

Input validation failures (do not retry):

Auth and rate limit failures:

Empty results: If fetch_url succeeds (no error thrown) but the returned Markdown content is very short or empty, this means the target page had no extractable main content (e.g., a page with only images, or a JavaScript-rendered SPA with no server-side HTML). In this case:

1. Try using a CSS selector to target a specific element

2. Try fetching with format: "html" to see the raw extracted HTML

3. The page may require JavaScript rendering (not currently supported)

Retry strategy for MCP clients:

• Retryable errors (502, 504, 429 rate limit): Retry up to 2 times with exponential backoff (1s, 3s)

• Non-retryable errors (401, 422, 429 quota): Do not retry. Fix the input or check configuration.

• Successful but empty: Try alternative selectors or inform the user the page has no extractable content

Error Handling

The MCP server returns clear error messages when something goes wrong:

REST API

The MCP server wraps the StripFeed REST API. You can also call the API directly:

# Single URL fetch (returns Markdown)
curl "https://www.stripfeed.dev/api/v1/fetch?url=https://news.ycombinator.com" \
  -H "Authorization: Bearer sf_live_your_key"

# With options: JSON format, CSS selector, max 500 tokens
curl "https://www.stripfeed.dev/api/v1/fetch?url=https://example.com&format=json&selector=article&max_tokens=500" \
  -H "Authorization: Bearer sf_live_your_key"

# Batch fetch (up to 10 URLs, with per-URL selectors)
curl -X POST "https://www.stripfeed.dev/api/v1/batch" \
  -H "Authorization: Bearer sf_live_your_key" \
  -H "Content-Type: application/json" \
  -d '{"urls": ["https://news.ycombinator.com", {"url": "https://docs.anthropic.com", "selector": "article"}]}'

# Check usage
curl "https://www.stripfeed.dev/api/v1/usage" \
  -H "Authorization: Bearer sf_live_your_key"

JSON response format (when format=json):

{
  "markdown": "# Page Title\n\nClean content here...",
  "html": "<article>...</article>",
  "text": "Plain text version...",
  "url": "https://example.com",
  "title": "Page Title",
  "tokens": 1250,
  "originalTokens": 14200,
  "savingsPercent": 91.2,
  "cached": false,
  "fetchMs": 342,
  "format": "json",
  "truncated": false,
  "selector": "article",
  "model": null
}

Response headers (included with every response):

Usage response:

{
  "plan": "pro",
  "usage": 1250,
  "limit": 100000,
  "remaining": 98750,
  "resetsAt": "2026-04-01T00:00:00.000Z"
}

SDKs

For programmatic access without MCP, use the official SDKs:

TypeScript/JavaScript:

import StripFeed from "stripfeed";

const sf = new StripFeed("sf_live_your_key");

// Fetch a URL
const result = await sf.fetch("https://example.com");
console.log(result.markdown);
console.log(`Tokens: ${result.tokens} (saved ${result.savingsPercent}%)`);

// With options
const article = await sf.fetch("https://example.com", {
  selector: "article",
  maxTokens: 500,
  ttl: 7200,
});

// Batch fetch
const batch = await sf.batch(["https://a.com", "https://b.com"]);

// Check usage
const usage = await sf.usage();
console.log(`${usage.usage} / ${usage.limit} requests used`);

npm: https://www.npmjs.com/package/stripfeed

Python:

from stripfeed import StripFeed

sf = StripFeed("sf_live_your_key")

# Fetch a URL
result = sf.fetch("https://example.com")
print(result["markdown"])
print(f"Tokens: {result['tokens']} (saved {result['savingsPercent']}%)")

# With options
article = sf.fetch("https://example.com", selector="article", max_tokens=500, ttl=7200)

# Batch fetch
batch = sf.batch(["https://a.com", "https://b.com"])

# Check usage
usage = sf.usage()
print(f"{usage['usage']} / {usage['limit']} requests used")

PyPI: https://pypi.org/project/stripfeed/

Pricing

The MCP server uses your StripFeed API key. Usage counts toward your plan limits:

Links

• API Docs

• OpenAPI Schema

• TypeScript SDK

• Python SDK

• GitHub

License

MIT