mcp-win-stdio-rag 0.2.5__tar.gz

mcp_win_stdio_rag-0.2.5/.gitignore

@@ -0,0 +1,63 @@
+__pycache__/
+*.py[cod]
+*$py.class
+*.so
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+*.manifest
+*.spec
+api.txt
+pypi-api.txt
+
+
+pip-log.txt
+pip-delete-this-directory.txt
+pypi api.txt
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+*.mo
+*.pot
+
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+.idea/
+.vscode/
+*.swp
+*.swo
+
+*.log

mcp_win_stdio_rag-0.2.5/PKG-INFO

@@ -0,0 +1,25 @@
+Metadata-Version: 2.5
+Name: mcp-win-stdio-rag
+Version: 0.2.5
+Summary: Windows-optimized RAG MCP Server with Playwright crawling, Link Graph trees, and local hybrid search.
+Author: Mohan Kumar Indala
+License-Expression: MIT
+Keywords: crawler,knowledge-graph,mcp,playwright,rag,stdio,vector-search
+Requires-Python: >=3.10
+Requires-Dist: beautifulsoup4>=4.12.0
+Requires-Dist: mcp-win-stdio>=0.2.5
+Requires-Dist: mcp>=1.2.0
+Requires-Dist: networkx>=3.0
+Requires-Dist: numpy>=1.24.0
+Requires-Dist: playwright>=1.40.0
+Description-Content-Type: text/markdown
+
+# mcp-win-stdio-rag
+
+Windows-optimized RAG MCP Server with Playwright crawling, Link Graph trees, and local hybrid search.
+
+## Features
+* Automated Async Playwright Crawler
+* NetworkX Link Tree Graph generation
+* SQLite-backed Hybrid Vector & BM25 search
+* Zero-token context cost cross-collection queries

mcp_win_stdio_rag-0.2.5/README.md

@@ -0,0 +1,9 @@
+# mcp-win-stdio-rag
+
+Windows-optimized RAG MCP Server with Playwright crawling, Link Graph trees, and local hybrid search.
+
+## Features
+* Automated Async Playwright Crawler
+* NetworkX Link Tree Graph generation
+* SQLite-backed Hybrid Vector & BM25 search
+* Zero-token context cost cross-collection queries

mcp_win_stdio_rag-0.2.5/pyproject.toml

@@ -0,0 +1,32 @@
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[project]
+name = "mcp-win-stdio-rag"
+version = "0.2.5"
+description = "Windows-optimized RAG MCP Server with Playwright crawling, Link Graph trees, and local hybrid search."
+readme = "README.md"
+requires-python = ">=3.10"
+license = "MIT"
+authors = [
+    { name = "Mohan Kumar Indala" }
+]
+keywords = ["mcp", "rag", "vector-search", "playwright", "crawler", "knowledge-graph", "stdio"]
+
+dependencies = [
+    "mcp-win-stdio>=0.2.5",
+    "mcp>=1.2.0",
+    "networkx>=3.0",
+    "numpy>=1.24.0",
+    "playwright>=1.40.0",
+    "beautifulsoup4>=4.12.0",
+]
+
+[project.scripts]
+mcp-win-stdio-rag = "mcp_win_stdio.rag.cli:main"
+mws-rag = "mcp_win_stdio.rag.cli:main"
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/mcp_win_stdio"]
+

mcp_win_stdio_rag-0.2.5/src/mcp_win_stdio/rag/__init__.py

@@ -0,0 +1,6 @@
+"""
+RAG MCP Server package for mcp-win-stdio.
+Automated Playwright Web Crawling, Link Graph Trees, and Local Hybrid RAG Search.
+"""
+
+__version__ = "0.2.5"

mcp_win_stdio_rag-0.2.5/src/mcp_win_stdio/rag/__main__.py

@@ -0,0 +1,5 @@
+#!/usr/bin/env python3
+from mcp_win_stdio.rag.cli import main
+
+if __name__ == "__main__":
+    main()

mcp_win_stdio_rag-0.2.5/src/mcp_win_stdio/rag/cli.py

@@ -0,0 +1,24 @@
+#!/usr/bin/env python3
+"""
+CLI entry point for mcp-win-stdio-rag.
+"""
+
+import argparse
+
+from mcp_win_stdio.rag.server import mcp
+
+
+def main():
+    parser = argparse.ArgumentParser(description="mcp-win-stdio-rag MCP Server")
+    parser.add_argument("--transport", default="stdio", choices=["stdio", "sse"], help="MCP transport mechanism")
+    parser.add_argument("--port", type=int, default=8000, help="Port for SSE transport")
+    args = parser.parse_args()
+
+    if args.transport == "sse":
+        mcp.run(transport="sse", port=args.port)
+    else:
+        mcp.run(transport="stdio")
+
+
+if __name__ == "__main__":
+    main()

mcp_win_stdio_rag-0.2.5/src/mcp_win_stdio/rag/crawler.py

@@ -0,0 +1,179 @@
+#!/usr/bin/env python3
+"""
+Playwright-powered Automated Section-Aware Documentation Crawler and Link Graph Builder.
+Supports Single-Page API docs with Anchor ID deep-links (e.g. Zenodo, Slate, Redoc).
+"""
+
+import hashlib
+from typing import Any, Dict, List, Optional, Set, Tuple
+from urllib.parse import urljoin, urlparse
+
+import networkx as nx
+from playwright.async_api import async_playwright
+
+
+def get_domain_hash(url: str) -> str:
+    """Generate a stable folder hash for a target URL domain."""
+    parsed = urlparse(url)
+    domain = parsed.netloc or parsed.path
+    return hashlib.md5(domain.encode("utf-8")).hexdigest()[:12]
+
+
+def normalize_canonical_url(url: str, base_url: str) -> str:
+    """Normalize URL, enforce HTTPS, strip tracking params and URL fragments."""
+    joined = urljoin(base_url, url)
+    parsed = urlparse(joined)
+    scheme = "https" if parsed.scheme in ("http", "https") else parsed.scheme
+    netloc = parsed.netloc.lower()
+    path = parsed.path.rstrip("/")
+    if not path:
+        path = "/"
+    return f"{scheme}://{netloc}{path}"
+
+
+class AsyncPlaywrightCrawler:
+    """Async Web Crawler with Section & Anchor-Aware Knowledge Graph Generation."""
+
+    def __init__(self, target_url: str, max_depth: int = 2, max_pages: int = 40):
+        self.target_url = target_url
+        self.max_depth = max_depth
+        self.max_pages = max_pages
+        self.domain = urlparse(target_url).netloc.lower()
+        self.visited_urls: Set[str] = set()
+        self.graph = nx.DiGraph()
+        self.pages_data: List[Dict[str, Any]] = []
+
+    def _is_same_domain(self, url: str) -> bool:
+        parsed = urlparse(url)
+        return parsed.netloc.lower() == self.domain or not parsed.netloc
+
+    async def crawl(self) -> Dict[str, Any]:
+        """Crawl target site and extract section-aware chunks and deep-link anchors."""
+        queue: List[Tuple[str, int, Optional[str]]] = [(self.target_url, 0, None)]
+        total_sections_count = 0
+
+        async with async_playwright() as p:
+            try:
+                browser = await p.chromium.launch(headless=True)
+            except Exception as e:
+                err_text = str(e)
+                if "Executable doesn't exist" in err_text or "playwright install" in err_text:
+                    import subprocess
+                    import sys
+
+                    sys.stderr.write("Playwright Chromium browser missing. Installing automatically...\n")
+                    sys.stderr.flush()
+                    subprocess.run([sys.executable, "-m", "playwright", "install", "chromium"], check=True)
+                    browser = await p.chromium.launch(headless=True)
+                else:
+                    raise
+            context = await browser.new_context(
+                user_agent="Mozilla/5.0 (Windows NT 10.0; Win64; x64) Antigravity-RAG-Crawler/1.0"
+            )
+            page = await context.new_page()
+
+            while queue and len(self.visited_urls) < self.max_pages:
+                current_url, depth, parent_url = queue.pop(0)
+                norm_url = normalize_canonical_url(current_url, self.target_url)
+
+                if norm_url in self.visited_urls or depth > self.max_depth:
+                    continue
+
+                self.visited_urls.add(norm_url)
+                self.graph.add_node(norm_url, depth=depth)
+
+                if parent_url:
+                    self.graph.add_edge(parent_url, norm_url)
+
+                try:
+                    await page.goto(norm_url, timeout=18000, wait_until="domcontentloaded")
+                    title = await page.title()
+
+                    # Extract all sections with their anchor IDs via JavaScript DOM traversal
+                    sections_data = await page.evaluate("""
+                        () => {
+                            const headings = Array.from(document.querySelectorAll('h1, h2, h3, h4, section[id], article[id]'));
+                            const results = [];
+                            
+                            if (headings.length === 0) {
+                                // Fallback for pages without standard headings
+                                results.push({
+                                    anchor_id: '',
+                                    section_title: document.title || 'Main',
+                                    text: document.body ? document.body.innerText.trim() : ''
+                                });
+                                return results;
+                            }
+
+                            for (let i = 0; i < headings.length; i++) {
+                                const current = headings[i];
+                                const anchor_id = current.id || current.getAttribute('name') || '';
+                                const section_title = current.innerText.trim();
+                                
+                                // Collect text content between this heading and the next heading
+                                let textContent = [];
+                                let nextNode = current.nextElementSibling;
+                                
+                                while (nextNode && !['H1','H2','H3','H4','SECTION','ARTICLE'].includes(nextNode.tagName)) {
+                                    if (nextNode.innerText && nextNode.innerText.trim()) {
+                                        textContent.push(nextNode.innerText.trim());
+                                    }
+                                    nextNode = nextNode.nextElementSibling;
+                                }
+
+                                const sectionBody = textContent.join('\\n\\n').trim();
+                                if (section_title || sectionBody) {
+                                    results.push({
+                                        anchor_id: anchor_id,
+                                        section_title: section_title,
+                                        text: sectionBody ? `${section_title}\\n\\n${sectionBody}` : section_title
+                                    });
+                                }
+                            }
+                            return results;
+                        }
+                    """)
+
+                    # Extract internal links for continued crawling
+                    hrefs = await page.eval_on_selector_all(
+                        "a[href]", "elements => elements.map(el => el.getAttribute('href'))"
+                    )
+
+                    total_sections_count += len(sections_data)
+                    page_info = {
+                        "url": norm_url,
+                        "title": title,
+                        "depth": depth,
+                        "parent": parent_url,
+                        "sections": sections_data,
+                        "sections_count": len(sections_data),
+                    }
+                    self.pages_data.append(page_info)
+
+                    # Enqueue child links
+                    if depth < self.max_depth:
+                        for href in hrefs:
+                            if not href or href.startswith(("javascript:", "mailto:", "tel:")) or href.startswith("#"):
+                                continue
+                            child_norm = normalize_canonical_url(href, norm_url)
+                            if self._is_same_domain(child_norm) and child_norm not in self.visited_urls:
+                                queue.append((child_norm, depth + 1, norm_url))
+
+                except Exception as e:
+                    self.pages_data.append(
+                        {"url": norm_url, "title": f"Error loading {norm_url}", "error": str(e), "sections": []}
+                    )
+
+            await browser.close()
+
+        tree_structure = nx.node_link_data(self.graph)
+        is_single_page = len(self.visited_urls) == 1 and total_sections_count > 5
+
+        return {
+            "target_url": self.target_url,
+            "total_pages_crawled": len(self.visited_urls),
+            "total_sections_extracted": total_sections_count,
+            "is_single_page_doc": is_single_page,
+            "pages": self.pages_data,
+            "site_graph": tree_structure,
+        }

mcp_win_stdio_rag-0.2.5/src/mcp_win_stdio/rag/guide.py

@@ -0,0 +1,47 @@
+"""
+Interactive and printable guide for the RAG MCP server.
+"""
+
+RAG_GUIDE = """
+# ================================================================
+# RAG MCP (MULTI-COLLECTION EMBEDDINGS & SEARCH) - USER GUIDE
+# ================================================================
+
+The RAG MCP server provides 9 specialized tools for multi-collection web crawling,
+incremental local codebase indexing, zero-clone remote GitHub repository streaming,
+token-compact hybrid search, and on-demand chunk expansion on Windows.
+
+----------------------------------------------------------------
+1. TOOL SUMMARY (9 TOOLS)
+----------------------------------------------------------------
+* Crawling & Ingestion:
+  - crawl_and_index_url(target_url, collection_name, max_depth, max_pages):
+    Asynchronous headless Playwright web crawler extracting clean markdown into vector collections.
+  - index_local_codebase(project_path, collection_name, include_code, extensions, force_reindex):
+    Fast AST & chunked indexing of local codebases with SHA-256 incremental cache.
+  - index_remote_repo(repo_url, collection_name, subpath, is_temp, auth_token):
+    Streams public or private GitHub repository archives directly in-memory without cloning.
+
+* Querying & Knowledge Retrieval (Agent-Friendly & Token-Efficient):
+  - query_knowledge_base(query, target_url_or_collection, top_k, max_chars_per_snippet, compact):
+    Performs hybrid BM25 + dense semantic vector search with relevance ranking.
+    Features compact snippet mode and character limits (default 500 chars) to prevent context window bloat.
+  - get_chunk_context(chunk_id, target_url_or_collection, window):
+    Retrieves full untruncated content for any chunk, optionally expanding surrounding context window.
+  - get_knowledge_tree(target_url_or_collection, filter_path, max_items):
+    Returns hierarchical outlines and indexed document paths in a collection with path filtering.
+
+* Collection Lifecycle & Cache Management:
+  - list_rag_collections():
+    Lists all indexed knowledge collections with document counts, chunk statistics, and sizes.
+  - delete_rag_collection(collection_name):
+    Permanently deletes a collection and its vector index from disk.
+  - prune_rag_cache(max_total_mb, purge_temp_only):
+    Frees disk space by deleting outdated crawl artifacts and orphan chunks.
+# ================================================================
+"""
+
+
+def print_rag_guide() -> None:
+    """Print the formatted RAG MCP guide to stdout."""
+    print(RAG_GUIDE.strip())