Skip to main content

​
Integrations

MutagenT integrations allow you to send traces from your LLM applications to the MutagenT platform for observability, analysis, and optimization.

​
How Integrations Work

Integrations use callback handlers that:
  1. Intercept LLM calls in your application
  2. Capture request/response data, timing, token usage
  3. Send traces to MutagenT asynchronously
  4. Don’t impact your application’s performance

​
Generate Integration Code

The easiest way to integrate is using the CLI: 
# Interactive - auto-detects your framework
mutagent integrate

# Direct framework selection
mutagent integrate mastra
mutagent integrate langchain
mutagent integrate vercel-ai
See CLI Integrations for full details.

​
Supported Frameworks

FrameworkPackageStatus
Mastra@mutagent/mastraAvailable
LangChain@mutagent/langchainAvailable
LangGraph@mutagent/langgraphAvailable
Vercel AI@mutagent/vercel-aiAvailable
GenericDirect APIAvailable

​
What Gets Tracked

LLM Calls

Input prompts, outputs, model used

Token Usage

Input/output tokens, costs

Latency

Request duration, time to first token

Errors

Failures, rate limits, timeouts

​
Integration Pattern

All MutagenT integrations follow a similar callback handler pattern: 
import { MutagentCallbackHandler } from '@mutagent/langchain';

// Create callback handler
const handler = new MutagentCallbackHandler({
  apiKey: process.env.MUTAGENT_API_KEY,
  // Optional: link traces to specific prompts
  promptId: 123,
});

// Add to your LLM framework
const chain = new ConversationChain({
  llm: new ChatOpenAI(),
  callbacks: [handler],
});

// Traces automatically sent to MutagenT
const response = await chain.invoke({ input: 'Hello!' });

​
Next Steps

LangChain

LangChain integration guide

LangGraph

LangGraph tracing guide

Mastra

Mastra integration guide