Leafly Strain Scraper

<div align="center">

</div>

<br>

Overview

This repository contains a collection of purpose-built Model Context Protocol servers, each designed around a specific capability: web scraping and structured data extraction, codebase navigation and analysis, LLM-powered text generation, and JSON querying. Every server exposes its functionality as MCP tools and resources, making them composable building blocks for AI agent workflows.

The servers span two language ecosystems — TypeScript for the Firecrawl integration and DeepSeek/JSON servers, Python for the codebase analysis server — and follow the MCP SDK conventions for tool definitions, resource URIs, and transport configuration (stdio and HTTP).

<br>

Technology Stack

<table>
<tr>
<td><strong>Languages</strong></td>
<td> </td>
</tr>
<tr>
<td><strong>MCP Framework</strong></td>
<td> </td>
</tr>
<tr>
<td><strong>Web Scraping</strong></td>
<td> </td>
</tr>
<tr>
<td><strong>AI Integration</strong></td>
<td> </td>
</tr>
<tr>
<td><strong>Data & Querying</strong></td>
<td> </td>
</tr>
<tr>
<td><strong>Runtime</strong></td>
<td> </td>
</tr>
<tr>
<td><strong>Tooling</strong></td>
<td> </td>
</tr>
</table>

<br>

Server Index

| Server | Language | Transport | Tools | Resources | Description |
|:-------|:---------|:----------|:------|:----------|:------------|
| Firecrawl Web Scraping | TypeScript | stdio | 10 | &mdash; | Web scraping, crawling, batch processing, LLM extraction, deep research, and cannabis strain data extraction |
| Codebase Analysis | Python | stdio | 6 | 4 | File system navigation, code search, project structure analysis, and real-time change monitoring |
| DeepSeek R1 | JavaScript | stdio | 5 | 5 | Text generation, summarization, streaming, multi-model support, and document processing via DeepSeek AI |
| JSON Manager | JavaScript | stdio / HTTP | 4 | 4 | JSONPath querying, advanced filtering, dataset comparison, and result caching |

<br>

---

Firecrawl Web Scraping Server

A comprehensive MCP server built on the Firecrawl platform for web scraping, content extraction, and structured data collection. Extends the base Firecrawl capabilities with a specialized cannabis strain data extraction pipeline that collects 66 standardized data points per strain from Leafly.com using dual extraction strategies: regex-based pattern matching and LLM-powered schema extraction.

Tools

| Tool | Description |
|:-----|:------------|
| firecrawl_scrape | Scrape a single page with format selection (markdown, HTML, screenshots), custom actions, and content filtering |
| firecrawl_map | Discover all URLs on a website and generate a site map |
| firecrawl_crawl | Recursively crawl a website with depth and page limits |
| firecrawl_batch_scrape | Scrape multiple URLs concurrently with queue-based processing |
| firecrawl_check_batch_status | Poll the status of an in-progress batch scrape job |
| firecrawl_check_crawl_status | Poll the status of an in-progress crawl job |
| firecrawl_search | Search the web and return scraped content from results |
| firecrawl_extract | LLM-powered structured data extraction using a caller-defined JSON schema |
| firecrawl_deep_research | Multi-step research workflow that scrapes, synthesizes, and reports on a topic |
| firecrawl_leafly_strain | Extract standardized cannabis strain data (cannabinoids, terpenes, effects, flavors, interactions) |

Strain Data Extraction Pipeline

The Leafly strain extractor is the most specialized component in this collection. It implements two complementary extraction strategies against the same data source:

Regex-based extraction parses raw HTML/markdown content with pattern-matching rules for cannabinoid percentages, terpene profiles, effect ratings, and flavor descriptors. This approach is deterministic and fast, but brittle against layout changes.

LLM-powered extraction uses Firecrawl's extract endpoint to send page content to an LLM with a structured JSON schema. This approach handles unstructured text, formatting variations, and missing data more gracefully, at the cost of API latency and token usage.

Both strategies normalize output to a consistent schema covering:

| Category | Fields |
|:---------|:-------|
| Cannabinoids | THC, CBD, CBG, CBN |
| Terpenes | Myrcene, Pinene, Caryophyllene, Limonene, Linalool, Terpinolene, Ocimene, Humulene |
| Medical Effects | Stress, Anxiety, Depression, Pain, Insomnia, Lack of Appetite, Nausea |
| User Effects | Happy, Euphoric, Creative, Relaxed, Uplifted, Energetic, Focused, Sleepy, Hungry, Talkative, Tingly, Giggly |
| Adverse Effects | Dry Mouth, Dry Eyes, Dizzy, Paranoid, Anxious |
| Flavors | Berry, Sweet, Earthy, Pungent, Pine, Vanilla, Minty, Skunky, Citrus, Spicy, Herbal, Diesel, Tropical, Fruity, Grape |
| Pharmacokinetics | Onset (minutes), Duration (hours) |
| Drug Interactions | Sedatives, Benzodiazepines, SSRIs, Opioid Analgesics, Anticonvulsants, Anticoagulants |

Normalization methodology: lab-tested data is prioritized. When exact values are unavailable, standardized normalization is applied (dominant terpene = 0.008, second = 0.005, third = 0.003). Effects and flavors are normalized to a 0.0&ndash;1.0 scale.

Quick Start

cd firecrawl-mcp-server
npm install
cp .env.example .env          # Add your FIRECRAWL_API_KEY
npm run build
npm start                     # Start the MCP server (stdio transport)

# CLI: extract strain data directly
npm run scrape-leafly -- output.csv "Blue Dream,OG Kush,Sour Diesel"

<br>

---

Codebase Analysis Server

…