/langgraph — LangGraph SDK Skill

/langgraph — LangGraph SDK Skill

You now have full access to the LangGraph SDK via MCP tools in the langchain-bridge server.

Available MCP tools

Tool	What it does
mcp__langchain-bridge__langgraph_run	Build + run a LangGraph StateGraph inline. Supports answer or plan_answer graph shapes.
mcp__langchain-bridge__langgraph_react_agent	Run a LangGraph ReAct agent with Tavily web search. Agent decides when to search.

When to use which

• langgraph_run(steps="answer") — single-node graph, one LLM call. Fastest.
• langgraph_run(steps="plan_answer") — two-node graph: planner → answerer. Better for complex tasks.
• langgraph_react_agent — multi-turn ReAct loop with web search. Use when live data needed.

Graph shapes in langgraph_run

steps="answer":
  START → [answerer] → END

steps="plan_answer":
  START → [planner] → [answerer] → END

Both use ChatGroq (llama-3.3-70b) as the LLM node.

Example invocations

User: /langgraph run a simple graph to explain quantum computing → langgraph_run(query="Explain quantum computing", steps="answer")

User: /langgraph plan and answer: what are the best practices for building RAG systems → langgraph_run(query="Best practices for RAG systems", steps="plan_answer")

User: /langgraph research latest LangGraph features with web search → langgraph_react_agent(query="Latest LangGraph features and updates 2025")