FetchV2 MCP Server
This server fetches web pages and returns clean, readable Markdown for LLMs and MCP clients.
fetch— retrieve a single webpage and extract its main content as Markdown (pagination viastart_index, raw HTML option, metadata, tables, links, robots.txt bypass).fetch_batch— fetch 2–10 URLs in one call and combine the results with---dividers.discover_links— list and regex-filter links on a page (up to 100, relative URLs resolved) to plan what to read.fetch_llms_txt— parse anllms.txtindex and optionally fetch all linked pages.Safety/quality features — private-network guard, 5 MB response cap, table and link preservation, and adjustable length limits to save context.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@FetchV2 MCP Server Fetch the Wikipedia article on MCP"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
FetchV2 MCP Server
FetchV2 is a Model Context Protocol (MCP) server that retrieves web pages and returns clean Markdown. It uses Trafilatura to remove navigation, advertisements, footers, and other page elements.
What it does
Tool | Use |
| Fetch one web page and extract its main content |
| Fetch up to 10 web pages in one request |
| Find and filter links on a web page |
| Read an llms.txt index and optionally fetch its linked pages |
FetchV2 can return raw HTML, preserve links and tables, and paginate long content.
The fetch tool checks robots.txt by default.
Related MCP server: singlefile-mcp
Quick start
Requirements
Install
uv.Install Python 3.11 or newer:
uv python install 3.11Install for Cursor or VS Code
Cursor | VS Code |
Configure another MCP client
Add this server definition to your MCP client configuration:
{
"mcpServers": {
"fetchv2": {
"command": "uvx",
"args": ["fetchv2-mcp-server@latest"]
}
}
}Common configuration file locations:
Claude Desktop on macOS:
~/Library/Application Support/Claude/claude_desktop_config.jsonClaude Desktop on Windows:
%APPDATA%\Claude\claude_desktop_config.jsonWindsurf:
~/.codeium/windsurf/mcp_config.jsonKiro:
.kiro/settings/mcp.jsonin your project
Install in a Python environment
Use one of these commands if you want to install the package directly:
uv add fetchv2-mcp-serverpip install fetchv2-mcp-serverTry it
Ask your MCP client to perform a task such as:
"Fetch the documentation from
<URL>.""Find links on
<docs URL>that containtutorial.""Read these pages and summarize their differences:
[url1, url2, url3]."
Typical documentation workflow
First, find the relevant pages:
discover_links(url="https://docs.example.com/", filter_pattern="/guide/")Then fetch the selected pages in one request:
fetch_batch(
urls=[
"https://docs.example.com/guide/intro",
"https://docs.example.com/guide/setup",
]
)Tool reference
fetch
Fetch one web page and extract its main content as Markdown.
fetch(
url: str,
max_length: int = 5000,
start_index: int = 0,
get_raw_html: bool = False,
include_metadata: bool = True,
include_tables: bool = True,
include_links: bool = False,
bypass_robots_txt: bool = False,
) -> strParameter | Type | Default | Description |
|
| required | Web page URL |
|
|
| Maximum number of characters to return |
|
|
| Character offset for pagination |
|
|
| Return raw HTML without extraction |
|
|
| Include the title, author, and date |
|
|
| Preserve tables in Markdown |
|
|
| Preserve links in Markdown |
|
|
| Skip the |
If the response is truncated, use the returned start_index value in the next call.
fetch_batch
Fetch up to 10 web pages and combine the results.
fetch_batch(
urls: list[str],
max_length_per_url: int = 2000,
get_raw_html: bool = False,
) -> strParameter | Type | Default | Description |
|
| required | Web page URLs to fetch |
|
|
| Maximum number of characters to return for each URL |
|
|
| Return raw HTML without extraction |
This tool reports a failed URL in its result and continues with the other URLs.
It does not check robots.txt.
discover_links
Find links on a web page and optionally filter them with a regular expression.
discover_links(url: str, filter_pattern: str = "") -> strParameter | Type | Default | Description |
|
| required | Web page URL to scan |
|
|
| Regular expression used to filter links |
The tool resolves relative links and returns up to 100 URLs.
fetch_llms_txt
Read an llms.txt file and list its documentation links.
fetch_llms_txt(
url: str,
include_content: bool = False,
max_length_per_url: int = 2000,
) -> strParameter | Type | Default | Description |
|
| required | URL of an |
|
|
| Fetch the content of all linked pages |
|
|
| Maximum number of characters to return for each linked page |
By default, this tool fetches only the llms.txt index.
Set include_content=True to fetch all linked pages.
This option can return a large response.
The tool resolves relative URLs, such as /docs/guide.md, against the llms.txt URL.
Security
FetchV2 fetches URLs that a model chooses. A web page can contain a prompt injection that asks the model to fetch an internal address. FetchV2 has these limits to reduce that risk.
Private network guard
FetchV2 refuses a request when the host is, or resolves to, a non-public address:
Loopback:
127.0.0.0/8,::1, andlocalhostnamesPrivate:
10.0.0.0/8,172.16.0.0/12,192.168.0.0/16, and IPv6 unique localfc00::/7Link-local:
169.254.0.0/16(including the cloud metadata address169.254.169.254) andfe80::/10Unspecified, multicast, shared (
100.64.0.0/10), and reserved rangesIPv4-mapped IPv6 addresses, such as
::ffff:127.0.0.1
The guard checks the first request and every redirect hop.
It also applies to the robots.txt request and to the pages that fetch_llms_txt loads.
To fetch local or private hosts on purpose, set this environment variable for the server:
{
"mcpServers": {
"fetchv2": {
"command": "uvx",
"args": ["fetchv2-mcp-server@latest"],
"env": { "FETCHV2_ALLOW_PRIVATE_NETWORKS": "1" }
}
}
}The values 1, true, and yes turn the guard off. The guard is on by default.
Response size limit
FetchV2 reads at most 5 MB (5,242,880 bytes) of each response body.
It refuses a response early when Content-Length is over the limit.
It also counts bytes while it reads, and it stops at the limit.
A cut body ends with a <!-- Truncated: response body exceeded ... --> comment.
Known limits
DNS rebinding. The guard resolves the host before the request, and then the HTTP client resolves it again to connect. A hostile DNS server can return a public address the first time and a private address the second time. The guard does not pin the connection to the checked address. If this risk matters, run the server where it cannot reach internal services, for example behind an egress firewall.
Proxies. If you set
HTTP_PROXYorHTTPS_PROXY, the proxy connects to the host. The proxy must apply its own rules.Untrusted content. Fetched page text is untrusted input. It can contain instructions written for the model. Do not let fetched text approve other tool calls. Be careful with
autoApprovefor this server.
Prompts
fetch_manualcreates a request to fetch and summarize one URL.research_topiccreates a request to research a topic with optional URLs.
Development
Clone the repository and install the development dependencies:
git clone https://github.com/praveenc/fetchv2-mcp-server.git
cd fetchv2-mcp-server
uv sync --devRun the tests:
uv run pytestRun the server with MCP Inspector:
uv run mcp dev src/fetchv2_mcp_server/server.pyRun lint and type checks:
uv run ruff check .
uv run pyrightContributing
Read CONTRIBUTING.md before you submit a change.
Support
Use the GitHub issue tracker to report a problem or request a feature.
License
This project uses the MIT License. See LICENSE for details.
Available Tools
4 toolsdiscover_linksA
Discover all links on a webpage. Use this BEFORE fetch_batch to find relevant URLs.
USE THIS TOOL WHEN:
Exploring a documentation site to find relevant pages
Building a list of URLs to fetch in batch
Finding all subpages under a section (e.g., all /api/ docs)
Checking what content exists before deciding what to read
RECOMMENDED WORKFLOW:
discover_links(url="https://docs.example.com/", filter_pattern="/guide/")
Review the returned links and select relevant ones
fetch_batch(urls=[selected_urls], max_length_per_url=1500)
FILTER EXAMPLES:
filter_pattern="/docs/" → Only links containing '/docs/'
filter_pattern="getting-started|quickstart" → Links with either term
filter_pattern=".md$" → Only markdown file links
NOTES:
Returns up to 100 links (more are noted but omitted to save context)
Relative URLs are automatically resolved to absolute URLs
JavaScript/mailto/anchor links are excluded
Inputs: url (required), filter_pattern. Returns: result
fetchA
Fetch a single webpage and extract its main content as clean markdown.
USE THIS TOOL WHEN:
You need to read an article, documentation page, or blog post
You want clean, readable text without boilerplate (navbars, ads, footers)
The user provides a specific URL to read
DO NOT USE WHEN:
You need to fetch multiple URLs (use fetch_batch instead - fewer round trips)
You want to discover what pages exist on a site (use discover_links first)
PAGINATION: Large pages are automatically truncated. The response will include 'use start_index=N to continue' - call again with that value to get more content.
EXAMPLES:
fetch(url="https://docs.example.com/guide") → Get documentation page
fetch(url="https://example.com/api", max_length=2000) → Shorter response to save context
fetch(url="https://example.com/data", include_tables=True) → Preserve tabular data
Inputs: url (required), max_length, start_index, get_raw_html, +4 more. Returns: result
fetch_batchA
Fetch multiple webpages in a single request and return combined content.
USE THIS TOOL WHEN:
You have 2-10 URLs to read (e.g., from discover_links results)
Comparing content across multiple pages
Gathering context from several documentation pages at once
KEY BENEFIT: One tool call instead of multiple fetch() calls = fewer round trips, faster results, and reduced overhead in supervised/approval workflows.
WORKFLOW EXAMPLE:
discover_links(url="https://docs.example.com", filter_pattern="/api/")
fetch_batch(urls=[...returned links...], max_length_per_url=1500)
NOTES:
Each URL's content is separated by '---' dividers
Failed URLs show inline errors without stopping other fetches
Robots.txt is NOT checked for batch fetches (assumes prior discovery)
Inputs: urls (required), get_raw_html, max_length_per_url. Returns: result
fetch_llms_txtA
Fetch and parse an llms.txt file to discover LLM-friendly documentation.
USE THIS TOOL WHEN:
A site provides an llms.txt file for AI-friendly content discovery
You need to understand what documentation is available
You want to fetch structured docs in a single request
WHAT IS llms.txt: A proposal for sites to provide LLM-friendly content at /llms.txt. It's a markdown file listing documentation links with descriptions. See https://llmstxt.org for the specification.
WORKFLOW:
fetch_llms_txt(url="https://example.com/llms.txt") → Get structure
Review sections and links
Either use include_content=True or fetch_batch for specific pages
EXAMPLES:
fetch_llms_txt(url="https://fastht.ml/docs/llms.txt")
fetch_llms_txt(url="https://example.com/llms.txt", include_content=True)
Inputs: url (required), include_content, max_length_per_url. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 4 tools
Each tool targets a distinct retrieval mode: discover_links maps a site, fetch retrieves one URL, fetch_batch retrieves multiple URLs, and fetch_llms_txt parses the llms.txt format. Descriptions explicitly clarify when to use fetch vs fetch_batch, so an agent should not misselect.
All tool names use a lowercase verb-first style: fetch, fetch_batch, fetch_llms_txt, discover_links. The only minor inconsistency is that plain 'fetch' lacks the object/scope qualifier the other names have, but the overall pattern remains predictable.
Four tools is well-scoped for a web fetching and content discovery server. Each tool fills a distinct role—single fetch, batch fetch, link discovery, and llms.txt support—without unnecessary bloat or redundancy.
The server covers the core workflow end-to-end: discover relevant links, then fetch them individually or in batches. Minor gaps exist, such as no dedicated search or robots.txt checking tool, but these do not create dead ends for the stated purpose.
Maintenance
Related MCP Connectors
Web scraping, browser automation, screenshots and AI data extraction for AI agents via MCP.
Web extraction MCP: page content, metadata, articles, contacts, email checks, URL probes.
Web scraping MCP server: any URL to Markdown or JSON, with JavaScript rendering and batch jobs.
MCP server for web extraction and rendering via AceDataCloud WebExtrator
Related MCP Servers
- AlicenseBqualityDmaintenanceA Model Context Protocol server that enables LLMs to fetch and process web content in multiple formats (HTML, JSON, Markdown, text) with automatic format detection.56MIT
- AlicenseNot gradedqualityDmaintenanceAn MCP server for intelligent web content extraction from JavaScript-heavy sites using single-file and trafilatura. It enables AI agents to fetch, render, and paginate through clean article content and metadata.16 PyPI23MIT
- AlicenseNot gradedqualityDmaintenanceMCP server that fetches raw HTML content from a given URL to provide web context to LLMs.1-
- AlicenseAqualityCmaintenanceMCP server for web page fetching (converting to Markdown/text with automatic fallback between Tavily and Firecrawl) and web search via Tavily.2MIT