Skip to main content
These guides are for developers whose agent is built with an agent framework. Each guide covers its supported MCP connection, tracing path, or both. Model routing is configured separately through Model Gateway.

Framework guides

LangGraph

Use MCP tools and trace LangChain callbacks.

Claude Agent SDK

Trace agent runs, model calls, and tool calls.

OpenAI Agents SDK

Connect MCP tools and a compatible OpenTelemetry instrumentor.

Google ADK

Connect MCP tools and export native spans.

CrewAI

Load MCP tools for crew tasks.

pi

Trace pi-agent-core with the TypeScript adapter.

Pydantic AI

Export Pydantic AI native OpenTelemetry spans.

Vercel AI SDK

Trace generation, streaming, and agent calls.

Model SDK guides

OpenAI SDK

Trace OpenAI model calls with one init() option.

Anthropic SDK

Trace Claude model calls with one init() option.

AWS Bedrock

Trace Bedrock Runtime model calls with one init() option.

Core offerings

Framework coverage

Look up the support level and recommended tracing path for frameworks and libraries without a guide.

Wrap unsupported framework

Trace a framework with no guide or instrumentation by wrapping its top-level call.

Claude Managed Agents

Trace agents that run in Claude’s hosted runtime.

Trace errors

Fix traces that land under the wrong identity or workspace, or don’t arrive.

Check connection

Run one agent request, flush tracing, and inspect the trace’s root and child spans. For MCP tools, confirm the agent lists and calls the tools its identity can use.

Choose where to start

  • Your framework has a guide? Open it from the table above.
  • No framework, just model calls? Use a model SDK guide above, or create spans with the atlanai SDK directly.
  • Want to see a run first? Trace first agent run records one run and opens its trace.

See also

  • atlanai SDK: Tracing tasks that apply to any framework.