claude-debugs-for-you MCP — Cas d'usage, Installer & Démo en direct

Pourquoi l'utiliser

Fonctionnalités clés

Language-agnostic — any VS Code-supported debugger works
Tools: set/remove breakpoints, continue, step, evaluate expressions, list stack
stdio or SSE transport from VS Code extension to MCP client
Works with Claude Desktop, Continue, Cursor
Requires a valid .vscode/launch.json

Démo en direct

Aperçu en pratique

claude-debugs-for-you.replay ▶ prêt

0/0

Installer

Choisissez votre client

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

{
  "mcpServers": {
    "claude-debugs-for-you": {
      "command": "npx",
      "args": [
        "-y",
        "claude-debugs-for-you"
      ],
      "_inferred": true
    }
  }
}

Ouvrez Claude Desktop → Settings → Developer → Edit Config. Redémarrez après avoir enregistré.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "claude-debugs-for-you": {
      "command": "npx",
      "args": [
        "-y",
        "claude-debugs-for-you"
      ],
      "_inferred": true
    }
  }
}

Cursor utilise le même schéma mcpServers que Claude Desktop. La config projet l'emporte sur la globale.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "claude-debugs-for-you": {
      "command": "npx",
      "args": [
        "-y",
        "claude-debugs-for-you"
      ],
      "_inferred": true
    }
  }
}

Cliquez sur l'icône MCP Servers dans la barre latérale Cline, puis "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "claude-debugs-for-you": {
      "command": "npx",
      "args": [
        "-y",
        "claude-debugs-for-you"
      ],
      "_inferred": true
    }
  }
}

Même format que Claude Desktop. Redémarrez Windsurf pour appliquer.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "claude-debugs-for-you",
      "command": "npx",
      "args": [
        "-y",
        "claude-debugs-for-you"
      ]
    }
  ]
}

Continue utilise un tableau d'objets serveur plutôt qu'une map.

~/.config/zed/settings.json

{
  "context_servers": {
    "claude-debugs-for-you": {
      "command": {
        "path": "npx",
        "args": [
          "-y",
          "claude-debugs-for-you"
        ]
      }
    }
  }
}

Ajoutez dans context_servers. Zed recharge à chaud à la sauvegarde.

claude mcp add claude-debugs-for-you -- npx -y claude-debugs-for-you

Une seule ligne. Vérifiez avec claude mcp list. Supprimez avec claude mcp remove.

Cas d'usage

Usages concrets : claude-debugs-for-you

Let Claude drive the debugger to narrow down a bug

👤 Devs stuck on a non-obvious bug ⏱ ~20 min intermediate

Quand l'utiliser : You've been reading logs for 20 minutes. Time to let Claude set strategic breakpoints and check variable state.

Prérequis

.vscode/launch.json with a runnable config — Use ${file} for the first config so Claude can run whichever file you have open
claude-debugs-for-you extension installed — VS Code Marketplace

Déroulement

Start the extension server

Start the debug MCP server. Note the path it prints and paste into Claude Desktop config.✓ Copié

→ Extension status bar shows running
Describe the bug

My function computeTotal() returns NaN when cart has sale items. Place a breakpoint at the start, step through, and tell me where it goes wrong.✓ Copié

→ Breakpoint set, program runs to it, Claude inspects locals
Diagnose

Based on the values at each step, what's the root cause and what's the smallest fix?✓ Copié

→ Root cause + patch

Résultat : Bug found by Claude driving the debugger — faster than print-logging.

Pièges

launch.json is misconfigured and the debugger can't start — Test with F5 in VS Code first; only then let Claude drive
Claude sets too many breakpoints and slows everything to a crawl — Tell it to use conditional breakpoints where possible

Combiner avec : git-2

Understand a complex runtime state via expression evaluation

👤 Devs reverse-engineering unfamiliar code ⏱ ~15 min intermediate

Quand l'utiliser : Code works but you don't know why. Set a breakpoint at a key line and ask Claude to explore the state.

Déroulement

Pause

Set a breakpoint at line X, run the test, and when paused tell me what's in scope.✓ Copié

→ Scope dump
Evaluate

Evaluate this.store.getState().users.filter(u => u.active).length. Then show me this.config.✓ Copié

→ Eval results

Résultat : Live exploration without writing test code.

Pièges

Eval with side effects mutates state — Stick to read-only expressions unless you intend mutation

Combinaisons

Associez-le à d'autres MCPs pour un effet X10

claude-debugs-for-you + git-2

Debug → fix → commit loop

Debug the failing test, apply the fix you propose, then commit with a message describing what was wrong.✓ Copié

Outils

Ce que ce MCP expose

Outil	Entrées	Quand appeler	Coût
set_breakpoint	file, line, condition?	Before/during a debug session	free
remove_breakpoint	id	Clean up between sessions	free
continue		Resume; returns next pause info	free
step_over		Single-step without entering functions	free
evaluate	expression: str	Inspect values at current pause point	free
get_stack		See call chain	free

Coût et limites

Coût d'exécution

Quota d'API: None
Tokens par appel: Small per call; aggregate over a session can be 10-50k if you explore a lot of state
Monétaire: Free, MIT
Astuce: Ask Claude to form a hypothesis first, then verify with targeted breakpoints — avoids aimless stepping.

Sécurité

Permissions, secrets, portée

Stockage des identifiants : None

Sortie de données : Local debugger; values shipped to your LLM provider

Ne jamais accorder : Arbitrary `evaluate` on production databases via a debugger session

Evaluation can have side effects depending on the expression — treat with the care you'd give eval().
Don't run against production processes; use reproducible local environments.