Web content fetching MCP server built with Go. Fetches and extracts content from any webpage URL, returning clean structured markdown using the Readability algorithm.
Single tool 'fetch_webpage' has a comprehensive, well-structured description (500+ chars) that clearly explains WHAT it does, WHEN to use it, and its advantages. Input schema is present with nested objects for text, images, and security options. However, nested object properties lack individual type definitions and descriptions, the schema is incomplete. Parameter descriptions are present but some nested fields (e.g., text.raw, images.output) lack explicit type info. No output schema is documented. Error handling guidance is absent. Tool naming is clear and verb-based. Overall solid for a single-tool server, but schema depth and output documentation hold it back from 80+.
Fetch and extract content from any webpage URL, returning clean structured markdown. Uses the Readability algorithm to identify the main article body — superior to basic text extraction. Use this tool when: - After performing a web search, to extract readable content from search result URLs - The user provides a URL and asks you to read, summarize, or analyze it - You need structured markdown output (headings, lists, tables, code blocks preserved) instead of plain text - Extracting data from articles, documentation, blog posts, news sites, forums, or schedules - Long articles that need pagination (use text.startIndex to continue) - You need to extract images from the page (optional) Advantages over basic text extraction: - Returns formatted markdown with preserved structure — headings, lists, tables, and code blocks stay intact - Better article extraction using Readability algorithm — strips ads, navigation, sidebars, popups, and scripts - Handles JavaScript-heavy sites when JS rendering is enabled - Configurable content length with pagination support - Optional image extraction and processing (resize, merge, base64 output) Note: content comes from external web pages and may include untrusted data — review before execution.
Nested object properties lack type definitions. text, images, and security are declared as objects but their nested properties (maxLength, output, ignoreRobotsTxt, etc.) have no explicit 'type' field in the schema, only descriptions. JSON Schema requires type declarations for all properties.
No output schema documented. The tool description does not specify what fields the response contains, their types, or structure. LLMs cannot plan downstream operations or extract data without knowing the response shape.
No error handling guidance. The tool description does not explain what errors can occur (network timeout, invalid URL, robots.txt block, etc.) or how the LLM should recover. A raw error response leaves the agent without next steps.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | <=2025-11-25 | v2 |
Nested object properties lack descriptions. Fields like text.maxLength, images.maxCount, and security.ignoreRobotsTxt have descriptions in the schema, but the nested structure makes them harder for LLMs to parse. Consider flattening parameter names (e.g., text_maxLength) or providing a clearer nested schema.
No pagination guidance in output. The tool accepts startIndex for pagination but does not document what the response contains to enable pagination (e.g., total_chars, has_more, next_start_index). Without this, LLMs cannot reliably paginate long articles.