The high-efficiency networking layer for LLMs. Reduce token consumption by 73–87% by cleaning web data before it hits your context window.
fetch-mcp demonstrates solid definition quality with well-structured schemas, comprehensive parameter descriptions, and clear tool purposes. All 6 tools have explicit input schemas with type constraints (min/max bounds, enums where appropriate). Descriptions are substantive (100-300 chars), explaining WHAT each tool does and WHEN to use it. Tool annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint) are correctly applied to all tools. However, output schemas are not formally documented, responses are described in prose but lack structured JSON Schema definitions. Error handling is present but generic (delegates to _handle_error). Parameter naming is verb-noun consistent (smart_fetch, web_search, css_query, browser_fetch, pdf_fetch, optimize_json), though some parameters like 'selector' and 'jsonpath' could benefit from more explicit type suffixes in descriptions.
Fetch a URL with a real browser (Playwright/Chrome), wait for JavaScript to execute, optionally extract via CSS selector, and return optimized markdown.
Fetch a page and return only the content matching a CSS selector.
Optimize a JSON object or array to reduce token usage. Aggressively prunes empty values, deduplicates repeated sub-objects, flattens deep nesting, and optionally returns a schema-first summary for large uniform arrays.
Fetch a PDF from a URL and extract its text content using pdfminer.six.
Fetch any URL and auto-optimize based on content type. For HTML: converts to clean markdown, stripping navigation, ads, and scripts. For JSON: returns a schema + sample by default for large arrays. Use the jsonpath parameter to drill into specific items or fields on follow-up calls. Dramatically reduces token usage compared to raw fetching.
Search the web using DuckDuckGo and return results as a markdown list.
Output schemas not formally documented. Tools return strings (markdown, JSON, or text) but lack structured response schema definitions. LLMs cannot plan downstream tool calls or extract specific fields without explicit output type documentation.
Error handling is generic. _handle_error() returns error messages but does not categorize errors as retryable, user-fixable, or fatal. LLMs receive no guidance on recovery strategy (retry vs ask user vs abort).
Parameter 'data' in optimize_json accepts 'JSON string, file path, or JSON-wrapped file reference' but description does not specify format constraints or examples. LLMs may pass invalid formats.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
No pagination support documented. smart_fetch and web_search return truncated results (max_chars limit) but do not expose offset/limit or next_cursor for iterating large datasets. Agents cannot reliably fetch all results.
optimize_json parameter 'data' description mentions 'file path' and 'JSON-wrapped file reference' but these are not standard MCP patterns. Unclear how file resolution works or what 'JSON-wrapped' means.