@modelcontextprotocol/server-fetch

Anthropic's official reference MCP server for fetching web pages and converting them to Markdown. Gives agents the ability to read any URL and get clean, LLM-friendly content.

Evaluated Mar 06, 2026 (0d ago) vcurrent
Homepage ↗ Repo ↗ Developer Tools fetch http web-scraping mcp anthropic official html markdown
⚙ Agent Friendliness
86
/ 100
Can an agent use this?
🔒 Security
81
/ 100
Is it safe for agents?
⚡ Reliability
82
/ 100
Does it work consistently?

Score Breakdown

⚙ Agent Friendliness

MCP Quality
88
Documentation
85
Error Messages
80
Auth Simplicity
100
Rate Limits
75

🔒 Security

TLS Enforcement
90
Auth Strength
70
Scope Granularity
70
Dep. Hygiene
88
Secret Handling
92

No credentials stored. Respects robots.txt (security/politeness). SSRF risk if allowed_hosts not configured — agents could be directed to fetch internal URLs. Configure allowed_hosts allowlist in production.

⚡ Reliability

Uptime/SLA
80
Version Stability
85
Breaking Changes
85
Error Recovery
78
AF Security Reliability

Best When

Your agent needs to read a specific URL. Fast, lightweight, converts HTML to Markdown automatically. Best for documentation, articles, and simple data pages.

Avoid When

Target site requires JavaScript rendering (use Puppeteer or Playwright MCP instead) or has aggressive bot protection.

Use Cases

  • Fetching web page content for research and fact-checking during agent reasoning
  • Reading API documentation pages that aren't in training data
  • Extracting structured data from HTML pages (pricing tables, specs, etc.)
  • Following links from search results to read full article content
  • Monitoring web pages for changes in agentic workflows

Not For

  • JavaScript-heavy SPAs that require browser execution (content won't render)
  • Sites with bot protection (Cloudflare, Imperva) that block simple HTTP fetches
  • High-frequency crawling (no rate limiting or politeness built in)

Interface

REST API
No
GraphQL
No
gRPC
No
MCP Server
Yes
SDK
No
Webhooks
No

Authentication

Methods: none
OAuth: No Scopes: No

No authentication required for the MCP server itself. Individual target URLs may require auth — not supported natively; use custom headers via fetch parameters.

Pricing

Model: open_source
Free tier: Yes
Requires CC: No

MIT licensed, free. Requires Python runtime. Uses httpx under the hood.

Agent Metadata

Pagination
none
Idempotent
Full
Retry Guidance
Not documented

Known Gotchas

  • JavaScript-rendered content not available — only static HTML is returned
  • Robots.txt is respected by default — some pages will be refused (set ignore_robots_txt=true to override)
  • Large pages are truncated at a configurable max_length — default 5MB; complex sites may be cut off
  • Cookies and sessions not maintained — each fetch is stateless; multi-page auth flows not supported
  • Some sites return different content to known bot user agents — detection possible even with custom User-Agent
  • Returns raw Markdown conversion of HTML — tables, complex layouts may not parse cleanly

Alternatives

Full Evaluation Report

Detailed scoring breakdown, competitive positioning, security analysis, and improvement recommendations for @modelcontextprotocol/server-fetch.

$99

Scores are editorial opinions as of 2026-03-06.

5215
Packages Evaluated
26151
Need Evaluation
173
Need Re-evaluation
Community Powered