Home / Docs / Introduction

DocShark

DocShark is a fast, local-first Model Context Protocol (MCP) server that scrapes, indexes, and serves documentation from any website. One package, unlimited doc sources, zero cloud dependency.

Context

Why this exists

Teams often end up with many disconnected MCP integrations. DocShark centralizes discovery, indexing, and retrieval so agents can work from a single source of truth.

Why DocShark?

AI coding assistants hallucinate documentation. Existing solutions all have gaps:

Solution	What it does	Gap
Context7	Pulls docs from GitHub repos	Gets source code, not rendered docs. Closed-source.
docs-mcp-server	Scrapes websites, indexes locally	Heavy deps (Playwright, LangChain, Node 22+).
Per-library MCPs	Serve one library each	One MCP per library doesn’t scale.

DocShark fills the gap: a single, general-purpose MCP that fetches docs from any website, stores them locally with zero-config search, and exposes them via the Model Context Protocol — all in one Bun-installed package.

Single server model

Replace many per-library MCP servers with one local index and one tool surface.

Agent-first outputs

Results are chunked and formatted so coding assistants can act on them immediately.

Rendered docs support

Targets real documentation pages instead of only repository markdown.

Lightweight runtime

Bun + SQLite + FTS5 keeps the stack compact without cloud dependencies.

Key Features

Any docs site — Point DocShark at any URL and it crawls, extracts, and indexes the content
SQLite + FTS5 — Blazing-fast full-text search with zero external dependencies
Local-first — Everything runs on your machine. No cloud, no API keys, no rate limits
Smart chunking — Heading-aware splitting preserves context and hierarchy
MCP + CLI — Full MCP server for AI assistants, plus a powerful command-line interface
Lightweight — ~375KB core footprint. No LangChain, no heavy browser deps by default

Quick Start

# Add documentation from any site
bunx docshark add https://svelte.dev/docs

# Search across your indexed docs
bunx docshark search "runes reactivity"

# Start the MCP server
bunx docshark start --stdio

Tech Stack

Component	Technology
Runtime	Bun
MCP Server	TMCP
Database	SQLite (bun:sqlite) + FTS5
Content Extraction	Readability.js + linkedom
Markdown Conversion	Turndown + GFM plugin
Validation	Valibot
CLI	cac

Next note

Getting Started

Install and configure DocShark as an MCP server for your AI coding assistant.