feat: add official MCP server for Claude Code / coding agents

[rajarshidattapy]

Problem

webpull is great as a CLI tool, but coding agents like Claude Code have no way to use it programmatically. The current workflow requires:

1. Manually running webpull <url> in the terminal
2. Hoping the agent can find and read the output directory
3. No way to trigger a pull from inside an agent session

This means agents can't answer "pull the Anthropic docs and find how streaming works" without manual intervention.

Proposed solution

Add an official mcp/ directory with a zero-dependency MCP server that exposes webpull as tools callable by any MCP-compatible coding agent (Claude Code, Cursor, etc.).

Tools exposed

Tool	Description
pull_docs	Crawl a URL using webpull's existing engine, save as markdown
search_docs	Full-text search across previously pulled docs
read_doc	Read a specific pulled file by path
list_docs	List all pulled doc sites and file counts
run_code	Execute JS/TS/Python/Bash in a sandboxed subprocess

Usage from Claude Code

# Agent can now do this in one shot:
"Pull https://docs.anthropic.com and find all streaming examples"
"Search my docs for tool_use"
"Run this Python snippet and show me the output"

Setup (one command)

claude mcp add webpull bunx webpull-mcp

Implementation

I have a working implementation ready as a single mcp/index.ts file — no npm dependencies beyond what webpull already uses (Bun runtime). Happy to open a PR if this direction looks good to you.

Why this belongs in the main repo

• webpull's primary use case is feeding docs to LLMs — MCP is the natural interface for that
• Keeps the MCP server versioned alongside the crawler so they don't drift
• One install (bun install -g webpull) gives you both the CLI and the agent interface