Integrations

Connect your production agent to Converra to diagnose failures, generate and test behavior fixes, and verify the approved change on live conversations.

Framework setup guides

Choose the framework your agent uses. Each guide includes installation, a copyable example, and the path from a failing conversation to a production verdict.

Framework	Start here
Vercel AI SDK	Wrap model calls from generateText and streamText
OpenAI Agents SDK	Send completed text conversations from Runner
LangGraph	Send conversation messages from graph state
Mastra	Send text conversations after agent.generate
Python	Send conversations with the official Python SDK

Use Converra MCP from your coding agent to diagnose the failure, test a proposed fix, and inspect its production verdict. You approve what goes live.

SDKs

Wrap your LLM clients with one line to automatically capture runs, enable A/B testing, and multi-agent tracing.

SDK	What it does	Status
Node.js SDK	converra.wrap(openai) — OpenAI, Anthropic, Vercel AI SDK, auto-instrumentation	Available
Python SDK	converra.wrap(openai) — OpenAI, Anthropic, LangChain callback handler	Available

Capability ladder:

• Diagnose — capture runs for insights and failure analysis
• Fix + Test — trigger optimizations with simulation testing
• Fix + Deploy — ship proven improvements automatically via webhooks or GitHub PR

Data Connectors

Import traces and feedback from your existing LLM observability platform.

Connector	Status
LangSmith	Available
Langfuse	Available
OpenTelemetry	Compatible via Traces API

If you already trace runs + feedback in your observability platform, Converra can import it directly.

Deployment

Integration	Description	Status
GitHub	Auto-PR when optimizations find winners — metrics, evidence, check runs, and merge-back sync	Available

Other Integrations

• MCP - Works with Cursor, Claude Code, Windsurf, and any MCP-compatible client
• REST API - Direct HTTP/JSON-RPC access