Debug, replay, and ship MCP and A2A agents.
Local-first docs for Agent Studio Open. Quickstart, concepts, MCP and A2A cookbooks, OTEL ingest, provider keys, CLI, editor, browser edition, architecture, privacy, sandbox, and troubleshooting.
No account required. Trace data stays on the machine until you export it.
stdio, SSE, HTTP, WebSocket. Agent cards. OpenTelemetry GenAI spans.
GitHub Actions, JUnit, PR comments, trace cards, intake packets.
Every launch-critical guide has a public route.
Pick the surface you are debugging. Each guide names the connection mode, what to inspect, what to record, and what artifact to export.
60-second quickstart
Connect a sample MCP server, record a tool call, replay it, and export a regression artifact.
Read guide →Concepts
Connections, sessions, replays, regression banks, trace cards, and intake packets.
Read guide →MCP cookbook
Worked MCP debugging recipes for stdio, SSE, HTTP, WebSocket, risk linting, and replay exports.
Read guide →A2A cookbook
Five worked examples for agent cards, capability negotiation, task lifecycle, and failure paths.
Read guide →OTEL ingest cookbook
Import OpenTelemetry GenAI spans, Phoenix JSON, OTLP HTTP JSON/proto, and OTLP gRPC into local traces.
Read guide →Provider keys
Configure OpenAI, Anthropic, Google, and Ollama without sending keys to AuraOne.
Read guide →CLI reference
The agentstudio binary for connect, record, replay, compare, risk-scan, A2A, and export.
Read guide →VS Code extension
Discover MCP servers from a workspace, inspect manifests, compose calls, and deep-link to desktop.
Read guide →Browser edition
Use Agent Studio Open from agentstudio.auraone.ai/web with remote transports and local browser storage.
Read guide →Architecture
Tauri shell, Rust core, Python sidecar, trace store, protocol adapters, and export pipeline.
Read guide →FAQ
Common product, protocol, privacy, commercial-roadmap, and launch questions.
Read guide →Privacy and telemetry
What stays local, what telemetry can be enabled, what exports contain, and how redaction works.
Read guide →Sandbox guide
Run untrusted stdio MCP servers with declared paths and limited network/file access.
Read guide →Troubleshooting
Connection failures, malformed manifests, Python sidecar issues, OTLP errors, and replay nondeterminism.
Read guide →The shortest path to a replay.
Install, connect to a sample server, record a session, replay it across providers, and export the artifact.
agentstudio --json connect stdio --command python --arg server.py agentstudio risk-scan ./server --format json --fail-on critical --out reports/risk.json agentstudio import-trace traces/openai-events.jsonl --format openai --store refund-path.ast agentstudio export gh-action regressions --out reports/agent-regression
Run the loop. Own the trace.
Local-first by design. The trace, the keys, the replay, and the export stay on the machine until the team decides otherwise.