Loading ToolBench

Reading the current corpus…

page2ai-mcp MCP server · C · ToolBench

page2ai-mcp

MCP server that converts any web page URL to clean Markdown for LLM context. Processing is local and SSRF-guarded; no external APIs, no telemetry.

2026-07-28+STDIO1 tools

64/100

Definition Quality81

Protocol Readiness45

Supportability47

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence · detected: stateful session

Deprecated protocol patterns detected

Summary

Single tool 'page_to_markdown' has a comprehensive, well-structured definition with clear naming, detailed description (280+ chars), and complete JSON Schema with all parameters typed and described. Input schema includes proper constraints (URI format, min/max for timeout_ms, boolean defaults). Output schema is not explicitly documented in the visible code, which is a notable gap. Error handling is mentioned in the description but not formalized in the tool definition. The tool follows verb_noun naming convention and accepts human-friendly input (URLs). No security issues detected, tool is read-only with SSRF guards and no credential parameters.

Tools (1)

page_to_markdownread onlysource verified81/100

Fetch a web page URL and convert it to clean Markdown optimized for LLM context. Preserves headings, code blocks (with language hints), links, and tables; strips ads, navigation, and cookie banners. For Mintlify docs (docs.anthropic.com, OpenAI platform, Vercel, Stripe, etc.) tries the `URL.md` convention first for cleanest output. Discovers tab groups statically and emits each panel as a `### Tab: {label}` section instead of concatenating (prevents Python+TypeScript examples merging into one broken block). Zero external API calls — parses locally via linkedom.

Issues (4)

HIGHtool ↗

Output schema not documented. Tool description does not specify the structure of returned Markdown (fields, format, metadata). LLMs cannot plan downstream processing without knowing what fields to extract.

page_to_markdown

HIGHrecovery-guide ↗

Error handling not formalized in tool definition. Description mentions 'blocks private ranges' and 'SSRF-guarded' but does not document what errors the tool returns, when they occur, or how to recover. LLMs cannot self-correct without error guidance.

page_to_markdown

MEDIUMpaginated-result ↗

No pagination or result-limiting mechanism documented. Tool accepts any URL and returns full Markdown. For very large pages (10MB cap mentioned in code but not in schema), no guidance on truncation or chunking strategy.

page_to_markdown

MEDIUMtool-description ↗

Parameter 'include_images' and 'include_frontmatter' are boolean flags with defaults, but no guidance on when to use them. Descriptions are minimal (under 50 chars each). LLMs may not understand the trade-offs.

Recommendations

Document the output schema explicitly: specify that the tool returns a JSON object with fields like 'markdown' (string), 'title' (string), 'url' (string), 'timestamp' (ISO 8601), 'language' (string), 'description' (string), and 'canonical_url' (string). Include example output.
Formalize error handling in the tool definition. Add a section documenting error cases: 'Invalid URL format', 'URL blocked (private range, loopback, cloud metadata)', 'Fetch timeout', 'Page too large (>10MB)', 'Invalid content type'. For each, provide recovery guidance.
Expand parameter descriptions for 'include_images' and 'include_frontmatter'. E.g., 'include_images: Include image references in Markdown. Disable to reduce token count for image-heavy pages. Default: true.' and 'include_frontmatter: Prepend YAML metadata (title, URL, timestamp, language, description, canonical URL). Useful for RAG pipelines; disable to get raw Markdown only. Default: true.'
Add a 'max_tokens' or 'truncate_at' parameter to allow LLMs to control output size. Alternatively, document the 10MB cap and explain truncation behavior in the description.
Include a 'prefer_markdown_convention' parameter (default true) to expose the Mintlify docs optimization. Describe when it applies (docs.anthropic.com, OpenAI platform, Vercel, Stripe, etc.) so LLMs know when to expect cleaner output.

Spec posture evidence

Inferred effective spec: 2026-07-28+.

Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained

Score history

Overall score trend

First recorded score · v2 rubric

64/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	C	64	2026-07-28+	v2