SmartFetch

What SmartFetch MCP server does

SmartFetch MCP server turns a public webpage URL into structured retrieval output for an agent. Results can include cleaned page text, Markdown, links, metadata, the final URL, HTTP status, word count, content hash, truncation status, elapsed time, and the method used to retrieve the page. The service can also report a request ID and service version.

The currently published Official MCP Registry release is V1.10.6. It provides the same retrieval engine through MCP tools and an HTTP API. HTTP access uses POST /fetch with a URL and optional max_chars and force_browser values. The response limit defaults to 20,000 characters, accepts at least 1,000, and caps at 50,000.

How it works

SmartFetch MCP server starts with ordinary HTTP retrieval. If that approach cannot obtain the needed page, it falls back to a real Chromium browser. Setting force_browser requests browser retrieval directly through the HTTP API. The retrieval engine includes request limits, rate limiting, concurrency controls, and SSRF protections; the provided material does not specify their exact configuration values.

The service supports x402 payment protection for live MCP tool execution and POST /fetch. In the published configuration, each execution costs $0.005 and uses exact payments. Base Sepolia is the default network; Base mainnet can use Coinbase's authenticated CDP facilitator when explicitly configured. Buyers can first submit a request without payment, read the PAYMENT-REQUIRED challenge, then retry with PAYMENT-SIGNATURE. Successful delivery produces PAYMENT-RESPONSE; failed responses with status 400 or higher are not settled.

Setup and configuration

The repository provides a native MCP server plus HTTP discovery and API routes. Free routes include /health, /meta, /docs, /openapi.json, /llms.txt, /.well-known/x402, /robots.txt, and /sitemap.xml. The API documentation and machine-readable metadata expose the active payment configuration.

Base Sepolia works with the default x402 facilitator without CDP credentials. Selecting Base mainnet requires CDP API credentials and supported facilitator configuration; startup fails closed when those requirements are invalid or missing. The material does not provide specific environment-variable names, so configuration should be taken from the repository's deployment instructions rather than inferred here.

Tools and capabilities

The published MCP surface contains four paid tools:

• fetch_webpage retrieves a webpage through the conditional HTTP/browser strategy.
• webpage_to_markdown returns webpage content as Markdown.
• extract_webpage_text returns cleaned webpage text.
• render_webpage uses the browser-rendering capability.

Each tool has its own MCP Bazaar declaration and payment resource while sharing the published $0.005 payment requirement. Runnable Python and TypeScript buyers are included for listing tools, enforcing a $0.005 maximum, handling the payment challenge and retry, and printing the result with a settlement receipt.

Limitations and notes

SmartFetch MCP server is intended for public webpages, not an arbitrary authenticated browsing session. The provided material does not document login handling or private-page access. Browser fallback also does not mean every page will be retrievable; HTTP failures can end in a typed diagnostic and an HTTP 502 response.

V1.11.0 adds conditional search_and_extract and extract_structured_data tools in the review branch, along with paid REST variants for search and extraction. That branch is not deployed, enabled, benchmark-approved, or published in the Official MCP Registry. Do not treat those capabilities as available in the current V1.10.6 release. Activity and access logging redact query strings and constrain diagnostic fields, but hosting-provider infrastructure logs may require separate controls.