@mpilot/langchain

LangChain JS adapter for @mpilot/tools. Converts the framework-agnostic Concierge registry into StructuredToolInterface[] ready for model.bindTools(tools) or any LangChain agent toolset.

Quickstart

import { getLangChainTools } from '@mpilot/langchain';

const tools = getLangChainTools(agent, [aaveTools /* …provider factories */]);
const bound = model.bindTools(tools);

getLangChainTools(agent, providerToolFactories?) mirrors createConciergeTools exactly: factories are composed, network-incompatible tools are filtered for agent.chainId, and omitting the factories yields an empty array. Seeing zero tools? Check that you passed the factories and that agent.chainId matches the networks your tools support.

Tool outputs are stringified (bigint-safe: wei amounts become decimal strings) so ToolMessage content is a deterministic string under the adapter's control — LangChain v1 would otherwise coerce objects itself. Parse the string if you need the structured value. Inputs are validated against the original Concierge Zod inputSchema (passed through by reference) before invoke runs.

Cancelling a LangChain run does not cancel an in-flight tool call — ConciergeTool.invoke takes no abort signal, so a started execution (e.g. an on-chain transaction) runs to completion.

@langchain/core is a peer dependency (^1.1.0); bring your own model — any LangChain chat model with tool-calling support works.