LangChain

langchain v1.4.6 HumanInTheLoopMiddleware can now pause once per tool call. Set interrupt_mode="per_call" and each gated call raises its own interrupt, which you answer by interrupt ID. Batched mode is still the default. Features Typed answers: each interrupt carries a response_schema listing the decisions that call allows, so a client can build its form from the schema. See Read the response schema. Checked before saving: LangGraph rejects an invalid answer before it is saved, so you can correct it and answer the same interrupt again. See Correct an invalid answer. Edits stay on the called tool: an edit can't switch tools, and its arguments are checked against the tool's own schema. Use per-call mode for new agents. LangChain plans to make it the default in the next major release. See Pause once per tool call and the human-in-the-loop overview.

langchain v1.4.0 MCP support now ships inside LangChain in the langchain.mcp namespace, built on FastMCP. It replaces the standalone langchain-mcp-adapters package. Install it with the mcp extra: Features One adapter, any transport: MCPAdapter infers the transport from its target — a URL, a local script over stdio, an in-process server, an MCPConfig dict for multiple servers, or a pre-built fastmcp.Client. await adapter.list_tools() returns LangChain tools ready for create_agent. See the MCP overview. Interrupt-driven elicitation: when a server asks for input mid-call, MCPAdapter surfaces the question as a LangGraph interrupt(), so a human answers and the run resumes. See Elicitation. Authentication through FastMCP: bearer tokens, full OAuth 2.1 with dynamic client registration, or any httpx.Auth, including per-user authentication in a deployment. See Authentication. Richer tool metadata: each tool carries its MCP provenance under an mcp namespace on its metadata, including annotations such as destructive_hint for gating destructive tools behind approval. See Tool metadata. Migrating Moving from langchain-mcp-adapters? MultiServerMCPClient is replaced by MCPAdapter, and several features have changed or been removed. See the migration guide.

langchain v1.3.0 This release adds support for version="v3" in stream_events / astream_events for langchain agents. Refer to the event streaming guide for details.

langchain-google-genai v4.0.0 We've re-written the Google GenAI integration to use Google's consolidated Generative AI SDK, which provides access to the Gemini API and Vertex AI Platform under the same interface. This includes minimal breaking changes as well as deprecated packages in langchain-google-vertexai. See the full release notes and migration guide for details.

langchain v1.2.0 create_agent: Simplified support for provider-specific tool parameters and definitions via a new extras attribute on tools. Examples: Provider-specific configuration such as Anthropic's programmatic tool calling and tool search. Built-in tools that are executed client-side, as supported by Anthropic, OpenAI, and other providers. Support for strict schema-adherence in agent response_format (see ProviderStrategy docs).

v1.2.0 langchain Structured output: Added ability to manually set strict mode when using providerStrategy for structured output. @langchain/openai New provider built-in tools: Support for file search, web search, code interpreter, image generation, computer use, shell, and MCP connector tools executed server-side by the provider. See Server-side tool use and the OpenAI chat integration. Content moderation: New moderateContent option on ChatOpenAI for detecting and handling unsafe content. Prefer responses API for GPT-5.2 Pro model. v1.3.0 @langchain/anthropic New provider built-in tools: Support for text editor, web fetch, computer use, tool search, and MCP toolset tools executed server-side by the provider. See Server-side tool use and the Anthropic chat integration. Exposed ChatAnthropicInput type for improved type safety. v1.1.0 @langchain/ollama Native structured outputs: Added support for native structured output via withStructuredOutput. Support for custom baseUrl configuration. v1.0.0 @langchain/community Jira document loader updated to use v3 API. LanceDB: Added similaritySearch() and similaritySearchWithScore() support. Elasticsearch hybrid search support. New GoogleCalendarDeleteTool. Various bug fixes for LlamaCppEmbeddings, PrismaVectorStore, IBM WatsonX, and security improvements. Other packages @langchain/xai: Native Live Search support. @langchain/tavily: Added Tavily's research endpoint. @langchain/mongodb: New MongoDB LLM cache. @langchain/mcp-adapters: Added onConnectionError option. @langchain/google-common: jsonSchema method support in withStructuredOutput. @langchain/core: Security fixes, better subgraph nesting in Mermaid graphs, UUID7 for run IDs.

v1.0.0 langchain Release notes Migration guide langgraph Release notes Migration guide

v1.0.0 langchain Release notes Migration guide langgraph Release notes Migration guide

langchain v1.1.0 Model profiles: Chat models now expose supported features and capabilities through a .profile attribute. These data are derived from models.dev, an open source project providing model capability data. Summarization middleware: Updated to support flexible trigger points using model profiles for context-aware summarization. Structured output: ProviderStrategy support (native structured output) can now be inferred from model profiles. SystemMessage for create_agent: Support for passing SystemMessage instances directly to create_agent's system_prompt parameter, enabling advanced features like cache control and structured content blocks. Model retry middleware: New middleware for automatically retrying failed model calls with configurable exponential backoff. Content moderation middleware: OpenAI content moderation middleware for detecting and handling unsafe content in agent interactions. Supports checking user input, model output, and tool results.

v1.1.0 Model profiles: Chat models now expose supported features and capabilities through a .profile getter. These data are derived from models.dev, an open source project providing model capability data. Model retry middleware: New middleware for automatically retrying failed model calls with configurable exponential backoff, improving agent reliability. Content moderation middleware: OpenAI content moderation middleware for detecting and handling unsafe content in agent interactions. Supports checking user input, model output, and tool results. Summarization middleware: Updated to support flexible trigger points using model profiles for context-aware summarization. Structured output: ProviderStrategy support (native structured output) can now be inferred from model profiles. SystemMessage for createAgent: Support for passing SystemMessage instances directly to createAgent's systemPrompt parameter and a new concat method for extending system messages. Enables advanced features like cache control and structured content blocks. Dynamic system prompt middleware: Return values from dynamicSystemPromptMiddleware are now purely additive. When returning a SystemMessage or string, they are merged with existing system messages rather than replacing them, making it easier to compose multiple middleware that modify the prompt. Compatibility improvements: Fixed error handling for Zod v4 validation errors in structured output and tool schemas, ensuring detailed error messages are properly displayed.