From install to a signed replay in sixty seconds.
Install the desktop app or open the browser edition. Connect to a sample MCP server. Record a tool call. Replay it. Export the regression artifact.
Local-first by default.
Desktop gets stdio MCP, the local OTLP receiver, OS keychain storage, sandbox mode, and signed update wiring from the shared Open Studio Platform shell.
The browser edition is intentionally smaller: remote SSE and HTTP MCP, A2A card testing, IndexedDB trace storage, and BYO provider keys protected by a passphrase.
brew install auraoneai/open/agent-studio-open agentstudio sample filesystem --open agentstudio connect --stdio "npx -y @modelcontextprotocol/server-filesystem ./sample-data"
Record, replay, and keep it.
Use Compose to call a tool. Press Record before sending. The session appears in Trace Browser with tool calls, model messages, latency, cost estimates, and risk annotations.
Replay with the same model to verify determinism. Export a GitHub Action or a JUnit report when the sequence should become release coverage.
Connect to a sample MCP server
Pick filesystem, fetch, or sqlite from the sample drawer. The connection panel shows transport, latency, and the manifest counts.
Confirm Manifest Inspector renders
Tools, Resources, Prompts, and JSON tabs all populate. Risk findings show up in the side rail.
Send a schema-backed tool call
Use Compose to call a tool with the form view, then switch to JSON to confirm the wire-level payload.
Replay and export a trace card
Replay the session against the same model to verify determinism, then export a trace card or a GitHub Action.
One command becomes a release gate.
The exported .ast trace is portable. Drop it into a GitHub Action, a JUnit pipeline, a PR comment, or an AuraOne intake packet. The next release has to clear it.
agentstudio record --session refund-path agentstudio replay refund-path.ast --compare anthropic,openai agentstudio export refund-path.ast --github-action --out .github/workflows/agentstudio.yml agentstudio export refund-path.ast --junit --out reports/agentstudio.xml
Next time, it's already a replay.
You shipped one trace. Repeat the loop on the path that actually broke. That is the regression bank.