📚
Agentic RAG LangGraph
blueprint · 4 agents · 1 MCP server
What it does
A retrieval pipeline that doesn't blindly trust its first search. It pulls context, grades whether the results actually answer the question, and if they don't, rewrites the query and tries again before generating. Loops until the context is good enough or it says it couldn't find the answer.
Depends on (MCP servers)
🔍search-mcpsearch ↗Agent flow
1 · Retriever
Pull candidate context
search: web
2 · Grader
Are these results relevant?
no tools · reasoning
3 · Rewriter
Rewrite the query if weak
no tools · reasoning
4 · Generator
Answer from good context
no tools · reasoning
fork & run
# fork into your project, wire your own servers
npx socketcat fork agentic-rag