Self-hosted MCP server that routes web scraping tools through your own HTTP or SOCKS5 proxy
Server has 4 tools with complete input schemas and reasonable descriptions. All tools are READ_ONLY and follow verb_noun naming (fetch_url, scrape_page, check_proxy, list_fingerprints). Schemas use Pydantic models with field validators and type hints. However, descriptions are terse (avg ~80 chars), lack WHEN/WHY guidance, and output schemas are not documented. Parameter descriptions exist but are minimal. No error recovery guidance. Missing tool annotations (readOnlyHint present implicitly via Risk field, but not in schema). No pagination support despite potential for large results.
Probe the proxy by calling an IP-echo endpoint. Returns the exit IP as seen by the public internet.
Fetch a URL through the configured proxy. Returns status, final URL, headers, and the response body as UTF-8 text (truncated to the configured limit).
List country codes for which a matching fingerprint preset is available.
Fetch a URL and extract fields using CSS selectors. If no selectors are provided, returns the page title and extracted text.
Output schemas not documented. LLMs cannot plan downstream calls or extract fields without knowing response structure. fetch_url returns {status, final_url, headers, body, truncated, bytes} but this is only visible in code, not in tool definition.
Descriptions lack WHEN/WHY context. 'Fetch a URL through the configured proxy' does not explain when to use fetch_url vs scrape_page, or what prerequisites exist (e.g., proxy must be configured). LLMs need explicit guidance for tool selection.
No error recovery guidance. fetch_url can return {error: 'blocked', reason: '...'} or {error: 'timeout'}, but tool description does not explain these failure modes or what the LLM should do next (retry? use different proxy? ask user?).
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | <=2025-11-25 | v2 |
list_fingerprints has empty input schema (no properties). Description is vague ('List country codes for which a matching fingerprint preset is available'). No output schema documented, LLM cannot know if it returns an array, object, or paginated result.
No tool annotations in schema. Risk field (READ_ONLY) is metadata, not part of JSON Schema. MCP spec supports readOnlyHint, destructiveHint, idempotentHint in tool definition, these should be explicit for LLM safety reasoning.