Graphify MCP — Anwendungsfälle, Installieren & Live-Demo

Warum nutzen

Hauptfunktionen

Multimodal input: code (25 languages via tree-sitter), PDF, Markdown, images, video, audio
Output: interactive HTML graph, graph.json, and Markdown report
Leiden community detection groups related concepts automatically
Slash commands: /graphify query, /graphify path, /graphify explain
Hooks integration so the graph is referenced automatically, not just on demand
--watch mode: auto-update on file changes; merge-graphs for cross-repo views

Live-Demo

In der Praxis

graphify-skill.replay ▶ bereit

0/0

Installieren

Wählen Sie Ihren Client

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "graphify-skill": {
      "command": "uv",
      "args": [
        "tool",
        "install",
        "graphifyy"
      ],
      "_inferred": true
    }
  }
}

Öffne Claude Desktop → Settings → Developer → Edit Config. Nach dem Speichern neu starten.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "graphify-skill": {
      "command": "uv",
      "args": [
        "tool",
        "install",
        "graphifyy"
      ],
      "_inferred": true
    }
  }
}

Cursor nutzt das gleiche mcpServers-Schema wie Claude Desktop. Projektkonfiguration schlägt die globale.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "graphify-skill": {
      "command": "uv",
      "args": [
        "tool",
        "install",
        "graphifyy"
      ],
      "_inferred": true
    }
  }
}

Klicken Sie auf das MCP-Servers-Symbol in der Cline-Seitenleiste, dann "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "graphify-skill": {
      "command": "uv",
      "args": [
        "tool",
        "install",
        "graphifyy"
      ],
      "_inferred": true
    }
  }
}

Gleiche Struktur wie Claude Desktop. Windsurf neu starten zum Übernehmen.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "graphify-skill",
      "command": "uv",
      "args": [
        "tool",
        "install",
        "graphifyy"
      ]
    }
  ]
}

Continue nutzt ein Array von Serverobjekten statt einer Map.

~/.config/zed/settings.json

{
  "context_servers": {
    "graphify-skill": {
      "command": {
        "path": "uv",
        "args": [
          "tool",
          "install",
          "graphifyy"
        ]
      }
    }
  }
}

In context_servers hinzufügen. Zed lädt beim Speichern neu.

claude mcp add graphify-skill -- uv tool install graphifyy

Einzeiler. Prüfen mit claude mcp list. Entfernen mit claude mcp remove.

Anwendungsfälle

Praxisnahe Nutzung: Graphify

Onboard to a 5-year-old legacy codebase in hours, not weeks

👤 New hires, contractors inheriting code ⏱ ~60 min intermediate

Wann einsetzen: You've joined a project with 10k+ files, dozens of internal services, and tribal knowledge spread across PRs and Confluence.

Voraussetzungen

uv installed — curl -LsSf https://astral.sh/uv/install.sh | sh
Graphify installed globally — uv tool install graphifyy && graphify install

Ablauf

Index the repo

/graphify index /workspace/monolith --include '**/*.{ts,py,md}'✓ Kopiert

→ Graph builds; report tells you cluster count and dominant concepts
Ask the graph

/graphify query 'how does payment processing connect to audit logging?'✓ Kopiert

→ Path through concepts with file references — not just a grep
Open the visual

/graphify explain 'user-account cluster' and open the HTML viewer✓ Kopiert

→ Interactive graph opens; cluster is highlighted with its bridges

Ergebnis: You know the conceptual shape of the codebase — where the hot paths live, which clusters are isolated, what depends on what.

Fallstricke

Graph becomes stale as the repo changes — graphify --watch to auto-update; or re-index before a big review session
Noise from vendored deps — Configure --exclude 'node_modules/,vendor/' on initial indexing

Kombinieren mit: filesystem · context7

Build a research graph over 200 papers and ask cross-paper questions

👤 Researchers, policy analysts, lit-review writers ⏱ ~90 min intermediate

Wann einsetzen: You have a folder of 200 PDFs and need to find cross-paper threads no single paper exposes.

Ablauf

Index the corpus

/graphify index /research/climate-pdfs✓ Kopiert

→ Concept clusters emerge (e.g. 'carbon pricing', 'adaptation finance')
Bridge-finding query

/graphify path 'adaptation finance' → 'sovereign debt'✓ Kopiert

→ Ordered chain of concepts + papers that connect the two
Generate a literature review

Summarize the 'adaptation finance → sovereign debt' bridge as a 400-word section with paper citations.✓ Kopiert

→ Draft paragraph with correct citations from the graph

Ergebnis: Insights that come from the structure of the corpus, not just keyword grep.

Fallstricke

Scanned PDFs don't OCR — Run OCR (ocrmypdf) before indexing; Graphify won't magic-text images

Kombinieren mit: filesystem

Merge graphs from multiple repos into one searchable view

👤 Platform teams owning microservice fleets ⏱ ~120 min advanced

Wann einsetzen: You run 15 services. You want one graph that spans them so an agent can follow a request end-to-end.

Ablauf

Index each service

For each repo under /services/, run graphify index with --out graph-<name>.json✓ Kopiert

→ 15 graph.json files on disk
Merge

graphify merge-graphs graph-*.json --out fleet.json✓ Kopiert

→ One unified graph with cross-repo edges
Ask end-to-end

/graphify query 'trace a POST /checkout through the fleet'✓ Kopiert

→ Path through multiple services, each node annotated with its repo

Ergebnis: Cross-service traceability without a service mesh or OTEL rollout.

Fallstricke

Same concept named differently across repos — Graphify's similarity merge handles most; for the rest, define aliases in the config

Kombinieren mit: filesystem · github

Kombinationen

Mit anderen MCPs für 10-fache Wirkung

graphify-skill + filesystem

Graphify points at files; filesystem lets the agent actually read/edit them

/graphify query 'retry logic' → list files, then filesystem read_text_file each one to draft a refactor.✓ Kopiert

graphify-skill + github

Turn graph queries into PR plans

Use the graph to find everywhere we do ad-hoc retries; open a GitHub issue listing each file with a proposed consolidation.✓ Kopiert

graphify-skill + sequential-thinking

Multi-step reasoning that uses the graph as an oracle

Plan the refactor step-by-step; at each step, consult /graphify query for impact.✓ Kopiert

Werkzeuge

Was dieses MCP bereitstellt

Werkzeug	Eingaben	Wann aufrufen	Kosten
/graphify index	path, include, exclude, --watch	First — build the graph	Claude calls for concept extraction (use prompt caching)
/graphify query	natural-language question	The primary query interface	free (local) + small model call
/graphify path	concept A → concept B	When you want to see how two ideas connect	free
/graphify explain	node/cluster id	Deep-dive on a single cluster	free + small model call
graphify merge-graphs	graph-*.json	Cross-repo / cross-folder views	free

Kosten & Limits

Was der Betrieb kostet

API-Kontingent: Local indexing is free; concept extraction uses Claude API (your key)
Tokens pro Aufruf: Indexing a 10k-file repo ≈ tens of dollars with caching; queries are cheap
Kosten in €: Free software; pay for the extraction calls only
Tipp: Cache aggressively — Graphify already does prompt caching for extraction; index once, query 1000x

Sicherheit

Rechte, Secrets, Reichweite

Credential-Speicherung: Uses your Claude API key for extraction; graph.json stored locally

Datenabfluss: Concept extraction calls go to Anthropic (your account). No telemetry from Graphify itself.

Niemals gewähren: Sharing graph.json publicly when indexed from a private repo — it embeds code snippets and filenames

graph.json can contain sensitive snippets. .gitignore it unless you want the graph shared with the repo.
--watch spawns file-system watchers; on huge repos, set include/exclude to avoid pegging a CPU.

Fehlerbehebung

Häufige Fehler und Lösungen

graphify: command not found after install

uv tool install graphifyy puts the binary in ~/.local/bin; ensure that's on PATH, or run with uv tool run graphify.

Prüfen: uv tool list | grep graphify

Indexing takes forever on a large repo

Scope with --include '**/*.{py,ts,md}' and --exclude 'node_modules/,dist/'. Concept extraction, not parsing, is the bottleneck.

Prüfen: Watch the progress log; parsing should be seconds, extraction takes the time

Queries return irrelevant clusters

Graph quality depends on the index — add README/docs files to the input set so Claude has project context for naming clusters

Prüfen: Rebuild including /docs/**/*.md; queries should sharpen

Merge-graphs produces duplicate nodes

Lower --similarity-threshold (default ~0.8) to 0.7 to merge near-duplicates more aggressively; tune per-repo

Alternativen

Graphify vs. andere

Alternative	Wann stattdessen	Kompromiss
Serena	You want LSP-precise symbol navigation and refactor tools, not conceptual clusters	No multi-modal input; no clustering
Memory MCP	You want Claude to remember session facts, not index a corpus	No graph; no query-by-concept
Raw embedding + vector DB	You need custom retrieval logic and don't care about community detection	You build the whole pipeline — graphify is opinionated and faster to ship

Mehr

Ressourcen

📖 Offizielle README auf GitHub lesen

🐙 Offene Issues ansehen

🔍 Alle 400+ MCP-Server und Skills durchsuchen