에이전트 강화
게시자가 제공하는 설치 방식입니다. PlugMCP 엔드포인트가 아닙니다.
원격 URL인가요, 내 컴퓨터에서 실행하는 명령인가요? 두 설치 방식의 차이
PlugMCP에서 실행
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
실행게시자 제공
Local-first knowledge graph for your browsing.
Weft turns browser tabs into a searchable, clustered knowledge graph using Louvain community detection and PageRank. No cloud. No accounts. Everything runs on your machine.
Think: Obsidian graph view, but for the web you already have open.
Local-first knowledge graph for browser tabs. Louvain clustering, PageRank, MCP server for AI assistants, Chrome extension, CLI. 100% private.
You have 80 tabs open. Some are related. Some are duplicates. You can't remember what you were researching yesterday.
Weft fixes this:
Live knowledge graph that tracks your browsing in real-time.
#keyword, and @domain filters with search historyj/k/Enter/o navigation, vim-stylechrome://extensionsextension/ folderBatch processing, terminal UI, API server, and MCP integration.
pip install weft-graph
Or from source:
git clone https://github.com/Avi-141/weft.git
cd weft
pip install -e .
# Build knowledge graph from browser tabs
weft weave
# Explore in terminal UI
weft explore
# Start API + MCP server
weft serve
weft weaveExtracts tabs from browsers and builds a knowledge graph.
# Default: all browsers, Louvain clustering
weft weave
# Chrome only, fast mode (no crawling)
weft weave --browser chrome --no-crawl
# With LLM summaries (requires Ollama)
weft weave --summarize
# Legacy clustering (Union-Find + mutual KNN)
weft weave --no-louvain
# Tighter clusters
weft weave --louvain-resolution 1.5
# Looser clusters
weft weave --louvain-resolution 0.7
weft exploreInteractive TUI for browsing your knowledge graph.
weft explore
weft explore my_graph.json
weft insightsPrint a markdown report of your browsing: top topics, key themes, domain breakdown.
weft insights
weft serveRun the API and MCP server for real-time extension sync and Claude Desktop integration.
weft serve
# Runs at http://localhost:8000
weft install-mcp (macOS)Configure Claude Desktop to talk to Weft via MCP.
weft install-mcp
Then ask Claude: "What was I researching about distributed systems?"
weft export-obsidianExport knowledge clusters as linked markdown notes to your Obsidian vault.
weft export-obsidian "/path/to/vault"
┌─────────────────────────────────────────────────────┐
│ DATA SOURCES │
│ Chrome (AppleScript) Firefox (sessionstore) │
│ Chrome Extension (live capture via chrome.tabs) │
└──────────────────────┬──────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────┐
│ EXTRACT + ANALYZE │
│ HTTP crawl (requests + trafilatura) │
│ Content script DOM extraction (extension) │
│ Keyword extraction (TF) + SimHash fingerprinting │
│ Optional: LLM summaries + embeddings (Ollama) │
└──────────────────────┬──────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────┐
│ DEDUPLICATE │
│ Canonical URL normalization (strip tracking, www) │
│ SimHash near-duplicate detection (same domain) │
│ Union-Find to merge duplicate sets │
└──────────────────────┬──────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────┐
│ CLUSTER (LOUVAIN + PAGERANK) │
│ Similarity matrix (Jaccard or cosine + domain) │
│ Louvain community detection (modularity Q) │
│ PageRank for node importance scoring │
│ PageRank-weighted TF-IDF group labeling │
└──────────────────────┬──────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────┐
│ GRAPH OUTPUT │
│ { tabs, groups, edges, stats } │
│ schema_version: 1 │
│ modularity score + clustering method in stats │
└──────┬───────────┬────────────┬─────────────────────┘
│ │ │
▼ ▼ ▼
CLI / TUI Extension Server / MCP
(Textual) (Cytoscape) (FastAPI)
Weft uses the Louvain method for clustering tabs. Instead of forcing tabs together by domain or arbitrary thresholds, Louvain optimizes modularity — finding groups where internal similarity is denser than expected by chance.
How it works:
Resolution parameter: Controls cluster granularity. Default is 1.0. Values >1.0 produce smaller, tighter clusters. Values <1.0 produce larger, looser ones.
Modularity Q: A score from -0.5 to 1.0 measuring cluster quality. Values above 0.3 indicate meaningful community structure. Weft reports this in graph stats and the extension UI.
Each tab gets a PageRank score based on its position in the similarity graph. Tabs that are highly connected to other important tabs rank higher.
Used for:
All scores are clamped to [0, 1].
Two-stage deduplication using Union-Find:
Weft exposes your browsing context to AI assistants via the Model Context Protocol.
Resources:
browsing://insights — browsing memory report with top topics, themes, sourcesbrowsing://groups — all clusters with labels, keywords, domains, sizesbrowsing://stats — graph metrics: tab count, modularity, clustering methodTools:
search_knowledge(query) — ranked search across titles, summaries, keywords, URLs. Scores by keyword overlap + PageRank.get_group_details(group_id) — cluster deep-dive: synthesized summary, tabs sorted by PageRank, top keywords and domainsget_tab_neighbors(tab_id) — graph neighborhood: all tabs connected by similarity or navigation edges, with edge weightsfind_recent_tabs(hours) — temporal query: tabs active within the last N hoursfind_stale_tabs(stale_hours) — cleanup candidates: tabs not accessed in N+ hours, sorted by stalenessfind_related_to_topic(topic) — topic expansion: finds direct keyword matches, then follows graph edges to discover related content you wouldn't find by search alone# Install MCP for Claude Desktop
weft install-mcp
# Start the server
weft serve
Example prompts for Claude:
# LLM summaries (embedding-based clustering)
brew install ollama
ollama pull llama3.1:8b
ollama pull nomic-embed-text
weft weave --summarize
# Firefox tab export
pip install lz4
# JS-heavy page rendering
pip install playwright
Weft is fully local:
MIT
Local-first knowledge graph for browser tabs. Louvain clustering, PageRank, MCP server for AI assistants, Chrome extension, CLI. 100% private.
도구
아직 도구 목록이 없습니다. 원격 서버는 첫 상태 확인 후에 도구 목록이 생깁니다.
FAQ
이 페이지에서 Cursor, Claude Desktop, Claude Code 또는 VS Code용 stdio 설정을 복사합니다. PlugMCP 프록시가 아니라 게시자의 설치 방식을 사용합니다.
이 레코드에는 필수 환경 변수가 등록되어 있지 않습니다.
Weft은(는) stdio를 통해 로컬에서 실행됩니다.
마지막 확인: 2026-10-28. Local-first knowledge graph for browser tabs. Louvain clustering, PageRank, MCP server for AI assistants, Chrome extension, CLI. 100% private.
