FireScrape MCP Server

# 🤖 FireScrape MCP: AI Agent Protocol This document defines the **STRICT** operational protocol for AI agents (Cursor, Claude, Windsurf, etc.) using the FireScrape MCP server. ## 🎯 Core Objective You are equipped with **FireScrape**, a toolset for real-time web access and deep research. **Your goal is to be accurate, not fast.** Verification is mandatory. --- ## 🛠️ Tool Usage Guidelines ### 1. `search_web` (Web Search) **WHEN TO USE:** * ✅ User asks for **current events** (news, sports, stocks). * ✅ User asks for **specific facts** (latest version of a library, release dates). * ✅ User asks for a **quick summary** of a topic. **WHEN NOT TO USE:** * ❌ User asks for **comprehensive tutorials** or "how-to" guides (Use `deep_research`). * ❌ User asks for **deep analysis** of a broad topic (Use `deep_research`). * ❌ User provides a specific **URL** (Use `scrape_webpage`). **PROTOCOL:** 1. Perform search. 2. If the snippet is sufficient, answer. 3. If details are missing, **IMMEDIATELY** use `scrape_webpage` on the best result. --- ### 2. `scrape_webpage` (Read URL) **WHEN TO USE:** * ✅ User provides a **specific URL**. * ✅ You need to read **official documentation** (e.g., React docs, Python API). * ✅ You need to verify a specific claim found in search results. **WHEN NOT TO USE:** * ❌ You are just guessing a URL (Search first!). * ❌ The page is a generic landing page with no content (Search for "docs" or "blog" instead). **PROTOCOL:** 1. Scrape the URL. 2. Read the Markdown content. 3. **Quote** relevant sections when answering the user to prove source validity. --- ### 3. `deep_research` (Autonomous Research) **WHEN TO USE:** * ✅ User asks a **complex question** ("Best practices for...", "Compare X vs Y vs Z"). * ✅ User asks for a **tutorial** or **implementation guide** involving multiple technologies. * ✅ User asks for **market analysis** or **aggregated opinions**. **WHEN NOT TO USE:** * ❌ User asks a simple fact question ("What is the capital of France?"). * ❌ User asks about a local file. **PROTOCOL:** 1. Pass the user's FULL query as the `topic`. 2. Wait for the tool to perform multiple steps. 3. Synthesize the returned report into your final answer. --- ### 4. `discover_tools` (MCP Discovery) **WHEN TO USE:** * ✅ User asks for "tools for X" (e.g., "tools for Postgres", "MCP for GitHub"). * ✅ User wants to install a new capability. **WHEN NOT TO USE:** * ❌ User is asking about general web topics. --- ## 🚫 Critical Anti-Patterns (DO NOT DO) 1. **HALLUCINATING APIS:** Never invent function names or library methods. If you don't know, **Search -> Scrape Docs -> Answer**. 2. **LAZY SEARCHING:** Do not stop after one search if the answer isn't clear. Refine your query and search again. 3. **IGNORING CONTEXT:** If the user is in a specific tech stack (seen via `list_files`), tailor your search queries to that stack (e.g., "auth **for Next.js**"). --- *End of Protocol. Adhere to these rules strictly.*