| name | langchain-docs-search |
| description | Search Langchain and LangGraph documentation using the langchain-docs MCP. Load when the user asks about Langchain/LangGraph APIs, concepts, or version-specific behavior you are not certain about. |
Langchain / LangGraph docs search
Use the langchain-docs MCP tools to look up accurate, version-specific information before answering questions about:
langchain, langchain-core, langchain-community APIs
langgraph graph construction, nodes, edges, state, checkpointers
langchain-openai, langchain-anthropic, langchain-ollama integrations
langchain-mcp-adapters and other ecosystem packages
Never hallucinate API signatures, parameter names, or class hierarchies. If uncertain, search first.
When to search
| Signal | Action |
|---|
| User asks "how do I use X in LangGraph" | Search before answering |
| User pastes a deprecation warning or error | Search the affected symbol |
| You're unsure about a parameter name/default | Search the class or function |
| User asks about a feature added after 2024 | Always search — training data may be stale |
Query patterns
Good queries are specific: library + symbol + what you need.
# Good
"LangGraph StateGraph add_node signature"
"langchain-core RunnableConfig configurable fields"
"LangGraph AsyncSqliteSaver from_conn_string"
"langchain-ollama ChatOllama streaming"
# Too vague
"langchain"
"how does langchain work"
Typical flow
- Call the docs search tool with a specific query.
- Read the returned excerpts — prefer the official API reference over guides when you need exact signatures.
- Cite the relevant doc section in your reply so the user can follow up.
- If the first result is insufficient, refine the query (add version, narrow the symbol name).
Do not use for
- General Python questions unrelated to the Langchain ecosystem.
- Questions the user has already answered with their own code snippet.
- Chinook / SQL questions — use
sql-agent for those.