← Explore all mcp servers

WEB RESEARCH / Model Context Protocol steering group

Fetch MCP reference server

Retrieve a URL as text or Markdown in chunks through a small local MCP server.

“Web content fetching and conversion for efficient LLM usage.”

01 / THE REASONING

Why this made the selection.

  • Provides direct URL retrieval without a search-service API account.
  • Chunked extraction and optional raw responses make long pages usable by a client.

02 / THE REVIEW RECORD

What we actually inspected.

Source review has boundaries.
A clear record is more useful than a “safe” badge.

Material inspected

  • src/fetch/README.md, pyproject.toml and server.py.
  • Robots checks, HTTP request handling, extraction and stdio startup.
  • Repository README and current licence transition.
  • src/fetch/LICENSE at the same commit.

Our findings

  • Autonomous fetch calls check robots.txt by default. The user-prompt path follows a separate code path.
  • The package exposes response offsets and raw output; no browser renderer or hosted search backend is configured.
  • Correction to the 2026-09-21 review: at d73f99ef the server had no private-address guard. At this commit server.py adds one (_refuse_private_destination) on the tool, the prompt and the robots.txt fetch, unless --allow-private-ips is passed.
  • The server registers one tool, fetch, with url, max_length, start_index and raw inputs, and one prompt, also named fetch. The README marks the tool readOnlyHint true and openWorldHint true.
  • src/fetch/LICENSE at this commit is the MIT licence (Copyright (c) 2024 Anthropic, PBC.) and pyproject.toml declares MIT, consistent with the root LICENSE transition text.

Not established by this review

  • Server launch, authenticated tool calls, and release-to-source parity were not tested.

The review applies to the material and revision named here. A newer upstream release can change its behavior.

03 / PUT IT TO WORK

Connect a server deliberately.

Upstream setup instructions ↗

DOCUMENTED COMMAND

pip install mcp-server-fetch

Copying does not execute this command. It may retrieve a newer version than the reviewed source.

  1. Install the package with pip, or configure the documented uvx subprocess in the MCP client.
  2. Use the Claude Desktop JSON example below; other clients use different configuration keys.
  3. Place the process in an appropriate network environment and only request sources you are allowed to access.

Before you start

  • Python tooling (pip or uv).
  • Outbound access to the requested sites; optional Node.js HTML simplification.

Documented configuration example

Use the client and authentication context described in the setup guidance. Configuration formats vary between hosts.

{
  "mcpServers": {
    "fetch": {
      "command": "uvx",
      "args": [
        "mcp-server-fetch"
      ]
    }
  }
}

04 / FOLLOW THE EVIDENCE

The source trail.

Our notes are separate from the original resource.
Check upstream before adopting a new version.

  1. https://raw.githubusercontent.com/modelcontextprotocol/servers/bbb701e7ac1e57f3edd1c2836b94c5457eaf7b9c/src/fetch/README.md

    Supports: summary, upstreamDescription, whySelected, bestFor, limitations, install, access, compatibility, transports, review, configExample, Private-address guard and --allow-private-ips, Proxy environment variables, Tool annotations

  2. https://raw.githubusercontent.com/modelcontextprotocol/servers/bbb701e7ac1e57f3edd1c2836b94c5457eaf7b9c/README.md

    Supports: publisher, upstreamDescription, limitations, review

  3. https://raw.githubusercontent.com/modelcontextprotocol/servers/bbb701e7ac1e57f3edd1c2836b94c5457eaf7b9c/LICENSE

    Supports: license, review

  4. Fetch Python package ↗Checked

    https://raw.githubusercontent.com/modelcontextprotocol/servers/bbb701e7ac1e57f3edd1c2836b94c5457eaf7b9c/src/fetch/pyproject.toml

    Supports: language, install

  5. https://raw.githubusercontent.com/modelcontextprotocol/servers/bbb701e7ac1e57f3edd1c2836b94c5457eaf7b9c/src/fetch/src/mcp_server_fetch/server.py

    Supports: review, limitations, access, transports, Private-address guard on tool, prompt and robots.txt fetch, One fetch tool and one fetch prompt

  6. https://raw.githubusercontent.com/modelcontextprotocol/servers/bbb701e7ac1e57f3edd1c2836b94c5457eaf7b9c/src/fetch/LICENSE

    Supports: license, MIT, Copyright (c) 2024 Anthropic, PBC.

Evidence & Ask