Sign in

Live traces for
AI agents.

Book a call

Private beta onboarding is hands-on. Connect Claude, let it provision observability, and debug agent plus backend traces in one session.

Live · receiving
trace_8f2a · live
98ms
agent.planner
openai
api
postgres
agent.writer
Latest span · now
Install anywhere
accepts sentry.envelope · otlp/http · otlp/json · agent.event
Hosted MCP beta

Provision to diagnosis.
Without leaving Claude.

AxonPush is not just another observability MCP. Your coding agent can set up its own telemetry, instrument the app, verify the first event, and investigate live backend-and-agent traces.

Book a call
$ claude mcp add --transport http
axonpush https://api.axonpush.xyz/mcp
  1. 1 Claude provisions an app, channels, environment, and scoped ingest key.
  2. 2 The AxonPush skill instruments the frameworks already in your repository.
  3. 3 Claude reads the newest combined trace and explains what failed.
Publish-only credentials. Read-only diagnosis tools.
One install, every agent

Available in
50+ AI coding agents.

Installs the integration as a skill your agent already knows how to follow. Detects your framework, creates an app and channel via our API, writes the integration code, prints a verify command.

Claude CodeCursorCodexOpenCodeClineGitHub CopilotWindsurfGemini+42 more
Claude Code/plugin
Everywhere elsenpx
What the skill does

Detects your framework — LangChain, CrewAI, Anthropic, OpenAI Agents, Vercel AI, Mastra, LangGraph, LlamaIndex, Google ADK, OTel — creates an app and channel via our API, writes the integration code into your project, and prints a verify command. Same flow whether you're in Claude Code, Cursor, Codex, or any of the 47 other supported agents.

LangChainCrewAIAnthropicOpenAI AgentsVercel AIMastraLangGraphLlamaIndexGoogle ADKOTel
The thing only axonpush does

One trace.
Agents and backend.

LangSmith stops at your agent graph. Datadog stops at your backend. axonpush walks the same trace across agents, LLM calls, HTTP hops, and database queries — because we ingest every source the same way.

trace_8f2a · 4 sources · 112ms
6 spans
agent.planner · start
anthropic · messages.create
api · GET /search
postgres · SELECT docs
agent.tool_call.end
agent.handoff → writer
agent llm http db
live
Wire it up manually

Or copy, paste, ship.

The SDKs are small. The integrations are one import.

$pip install axonpushdocs
1from axonpush import AxonPush
2from axonpush.integrations.langchain import AxonPushCallbackHandler
3
4client = AxonPush(api_key="ak_...")
5
6llm.invoke(
7    "Plan a trip to Kyoto",
8    callbacks=[AxonPushCallbackHandler(client)],
9)
10# Every chain, tool, and LLM token now streams to axonpush.
Also shipped

Every tool call, in order.

Sub-second SSE fan-out. No polling, no batch lag. Agent events, backend logs, and OTel spans land on the same timeline while the agent is still running.

Already ship OTel? One env var.

OTLP/HTTP and a JSON normalizer share the same endpoint. Point your existing collector — or your Sentry DSN — at api.axonpush.xyz. No code change.