@mrclrchtr/supi-web

Fetch web pages as clean Markdown and query library documentation via Context7 for the pi coding agent.

Install

Included in @mrclrchtr/supi, or install standalone:

pi install npm:@mrclrchtr/supi-web

For local development:

pi install ./packages/supi-web

After editing the source, run /reload to pick up changes.

Package surfaces

• @mrclrchtr/supi-web/api — programmatic exports (conversion helpers, fetch utilities, tool factories)
• @mrclrchtr/supi-web/extension — aggregated pi extension entrypoint that registers all three tools

pi.extensions still points at the real file path ./src/extension.ts inside the package. The /api and /extension paths are consumer-facing package exports, not manifest aliases.

What it adds

Registers three agent-callable tools:

Tool	Purpose
web_fetch_md	Fetch an http(s) URL and return clean Markdown
web_docs_search	Search Context7 for libraries by name
web_docs_fetch	Retrieve up-to-date documentation for a specific library via Context7

web_fetch_md — Web Page to Markdown

Parameters

Parameter	Type	Default	Description
url	string	required	http:// or https:// URL to fetch
output_mode	"auto" | "inline" | "file"	"auto"	auto returns inline if ≤15,000 chars, else writes to temp file
abs_links	boolean	true	Absolutize relative links and image sources
timeout_ms	number	30000	Fetch timeout in milliseconds

Content negotiation

The tool tries multiple strategies to get clean Markdown:

1. HEAD negotiation — checks content-type for Markdown
2. Sniffing — range GET of first 8KB to detect Markdown/plain text vs HTML
3. Sibling probing — tries .md / .markdown variants (e.g. page.md for page.html)
4. HTML conversion — full GET → JSDOM + Readability + Turndown → clean Markdown

web_docs_search — Library Lookup

Searches Context7's library index and returns a Markdown table of matching libraries with metadata.

Parameters

Parameter	Type	Default	Description
library_name	string	required	Library name to search for (e.g. "react", "next.js", "fastapi")
query	string	required	What the agent is trying to do — used for relevance ranking

Output

Returns a Markdown table with columns: Name, ID, Description, Trust Score, Benchmark Score, Snippet Count, Versions. The agent picks a library ID from these results to pass to web_docs_fetch.

web_docs_fetch — Documentation Retrieval

Fetches up-to-date documentation context for a specific library via Context7's API.

Parameters

Parameter	Type	Default	Description
library_id	string	required	Context7 library ID (e.g. /facebook/react, /vercel/next.js/v15.1.8)
query	string	required	Specific question about the library
raw	boolean	false	When true, returns JSON-serialized snippet objects instead of plain text Markdown

API Key (optional)

The tool reads the CONTEXT7_API_KEY environment variable automatically when set. Without a key, it works with lower rate limits. Get a key at context7.com/dashboard.

Source files

• src/api.ts — public package exports
• src/extension.ts — aggregated extension entrypoint
• src/web.ts — web_fetch_md tool registration
• src/docs.ts — web_docs_search + web_docs_fetch tool registration
• src/context7-client.ts — thin wrapper around @upstash/context7-sdk
• src/fetch.ts — HTTP fetch logic
• src/convert.ts — HTML to Markdown conversion

Commands

pnpm vitest run packages/supi-web/
pnpm exec tsc --noEmit -p packages/supi-web/tsconfig.json
pnpm exec biome check packages/supi-web/