/ Annuaire / Playground / langsmith-fetch-skill
● Communauté OthmanAdi ⚡ Instantané

langsmith-fetch-skill

par OthmanAdi · OthmanAdi/langsmith-fetch-skill

Debug LangChain / LangGraph agents from your terminal — fetches LangSmith Studio traces into Claude Code and analyzes failures, latency, tool calls.

A focused observability skill. When you ask Claude to debug an agent, it runs langsmith-fetch against LangSmith Studio, pulls trace data, analyzes execution patterns, and surfaces failures, root causes, latency hotspots, and tool-call issues. Export sessions for later review.

Pourquoi l'utiliser

Fonctionnalités clés

Démo en direct

Aperçu en pratique

langsmith-fetch-skill.replay ▶ prêt
0/0

Installer

Choisissez votre client

~/Library/Application Support/Claude/claude_desktop_config.json  · Windows: %APPDATA%\Claude\claude_desktop_config.json
{
  "mcpServers": {
    "langsmith-fetch-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/OthmanAdi/langsmith-fetch-skill",
        "~/.claude/skills/langsmith-fetch-skill"
      ],
      "_inferred": true
    }
  }
}

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

~/.cursor/mcp.json · .cursor/mcp.json
{
  "mcpServers": {
    "langsmith-fetch-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/OthmanAdi/langsmith-fetch-skill",
        "~/.claude/skills/langsmith-fetch-skill"
      ],
      "_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": {
    "langsmith-fetch-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/OthmanAdi/langsmith-fetch-skill",
        "~/.claude/skills/langsmith-fetch-skill"
      ],
      "_inferred": true
    }
  }
}

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

~/.codeium/windsurf/mcp_config.json
{
  "mcpServers": {
    "langsmith-fetch-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/OthmanAdi/langsmith-fetch-skill",
        "~/.claude/skills/langsmith-fetch-skill"
      ],
      "_inferred": true
    }
  }
}

Même format que Claude Desktop. Redémarrez Windsurf pour appliquer.

~/.continue/config.json
{
  "mcpServers": [
    {
      "name": "langsmith-fetch-skill",
      "command": "git",
      "args": [
        "clone",
        "https://github.com/OthmanAdi/langsmith-fetch-skill",
        "~/.claude/skills/langsmith-fetch-skill"
      ]
    }
  ]
}

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

~/.config/zed/settings.json
{
  "context_servers": {
    "langsmith-fetch-skill": {
      "command": {
        "path": "git",
        "args": [
          "clone",
          "https://github.com/OthmanAdi/langsmith-fetch-skill",
          "~/.claude/skills/langsmith-fetch-skill"
        ]
      }
    }
  }
}

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

claude mcp add langsmith-fetch-skill -- git clone https://github.com/OthmanAdi/langsmith-fetch-skill ~/.claude/skills/langsmith-fetch-skill

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

Cas d'usage

Usages concrets : langsmith-fetch-skill

Diagnose a slow agent without opening LangSmith UI

👤 LangChain/LangGraph devs ⏱ ~20 min intermediate

Quand l'utiliser : An agent that used to be snappy is now slow; you want root cause.

Prérequis
  • LangSmith API key — smith.langchain.com -> Settings -> API Keys
  • Skill installed — git clone https://github.com/OthmanAdi/langsmith-fetch-skill ~/.claude/skills/langsmith-fetch-skill
Déroulement
  1. Ask Claude to diagnose
    Why is my agent <name> slow? Pull the last 20 traces from LangSmith and find the hotspot.✓ Copié
    → Skill fetches traces; reports latency by step
  2. Drill in
    Focus on trace <id> — break down tool-call times.✓ Copié
    → Per-tool latency with the outlier called out

Résultat : A specific step to optimize, not a hunch.

Pièges
  • Optimizing the average when tail latency is the problem — Ask for p95/p99 specifically
Combiner avec : filesystem

Root-cause an agent failure from a trace id

👤 Anyone running LangGraph in production ⏱ ~15 min intermediate

Quand l'utiliser : A user reported an error; you have the trace id.

Déroulement
  1. Fetch + analyze
    Debug trace <id>. What was the immediate error, what's the root cause in the graph?✓ Copié
    → Error step + upstream decision that led to it

Résultat : A fix targeting the actual decision point, not the symptom.

Find which node burns the most tokens

👤 Teams optimizing LLM spend ⏱ ~30 min intermediate

Quand l'utiliser : Monthly review of agent spend.

Déroulement
  1. Aggregate token usage
    Across last 7 days' traces, rank graph nodes by total token usage.✓ Copié
    → Ranked table with % share
  2. Propose tightening
    For the top 3, propose prompt-compression or model downgrade options.✓ Copié
    → Targeted opt list

Résultat : Measurable spend reduction without chasing tiny savings.

Combinaisons

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

langsmith-fetch-skill + filesystem

Save exports alongside the agent code for lineage

Export the debugging session to debugging/traces/<date>.json.✓ Copié
langsmith-fetch-skill + github

Attach trace summary to a bug report

Summarize trace <id> and post as comment on GitHub issue #42.✓ Copié

Outils

Ce que ce MCP expose

OutilEntréesQuand appelerCoût
fetch_recent project, limit Broad diagnosis LangSmith quota
fetch_trace trace_id Deep dive on one run LangSmith quota
analyze_failures traces Post-fetch analysis 0
export_session traces Archive for later 0

Coût et limites

Coût d'exécution

Quota d'API
LangSmith: depends on plan; Studio has generous quotas
Tokens par appel
Low — traces are structured
Monétaire
Free skill; LangSmith free tier covers most devs
Astuce
Use fetch_recent with a narrow window, not full history

Sécurité

Permissions, secrets, portée

Portées minimales : LangSmith: read on your project(s)
Stockage des identifiants : LANGCHAIN_API_KEY in env
Sortie de données : Only to LangSmith (where the traces already live)
Ne jamais accorder : Write to LangSmith — skill only reads

Dépannage

Erreurs courantes et correctifs

langsmith-fetch: command not found

Install LangSmith CLI or ensure it's on PATH

Vérifier : which langsmith-fetch
Empty results

Check project name and time window; traces are project-scoped

Alternatives

langsmith-fetch-skill vs autres

AlternativeQuand l'utiliserCompromis
LangSmith web UIYou prefer clickingNo Claude analysis

Plus

Ressources

📖 Lire le README officiel sur GitHub

🐙 Voir les issues ouvertes

🔍 Parcourir les 400+ serveurs MCP et Skills