Use Agent Studio from your AI coding tool

Status: Current | Last reviewed: 2026-09-29

Claude Code and Codex can connect to your Agent Studio tenant. From your coding tool you can explore agents and workflows, edit drafts, use Generate with AI, ask the Studio assistant questions, and investigate recent conversations or workflow runs.

Your browser handles authentication through the same SSO login you use for Agent Studio. You do not need to copy a personal token into Claude or Codex.

Direct edits remain drafts. Nothing becomes live unless you explicitly ask the coding tool to publish and approve the publish action.

1. Configure the CLI

Install the trueserve binary provided by your administrator, then run:

trueserve configure

Enter the same public Agent Studio address you use in the browser. The optional tsf_* token prompt is only for the separate trueserve test command; press Enter to skip it if you only need coding-tool access.

2. Connect Claude Code or Codex

trueserve mcp install --tool both
# or use --tool claude / --tool codex

Then authenticate:

Your browser opens Agent Studio. Sign in through SSO if needed. The authorization screen shows the active tenant and the access requested by the coding tool. Review both, then choose Authorize. Choose Deny if the tenant or access does not match what you intended.

Your administrator controls which capabilities your role can approve. A coding tool may receive any combination of read, draft editing, publishing, debug, and Studio AI access for agents or workflows. Missing tools usually mean the corresponding access was not approved or is not part of your role.

3. Explore, build, generate, or ask

Use natural requests in your coding tool, for example:

Studio AI question sessions are advisory and cannot change a draft. If you ask for a change, the coding tool uses the separate validated authoring flow.

For direct editing, the skill starts from the current draft, discovers valid services and tools, validates its changes, and saves only if the draft has not changed since it was read. If you or a collaborator save in Studio first, the coding tool must re-read and merge its intended change rather than overwriting the newer draft.

Generate with AI runs on the Agent Studio generation service. Your coding tool can report progress and questions while it runs. Direct authoring instead uses the model supplied by Claude Code or Codex.

For scenario builds, the coding tool can read the same scenario-builder skill used by Studio and plan checks for authentication links, human review, voice wrap-up, visual components and Copilot latency. It can also discover native capabilities and published component/card contracts. These planning tools do not run tests, publish components or grant access to operations missing from your connection.

4. Publish

Publishing affects callers and embeds. The coding tool must receive an explicit publish request and use a confirmed publish action. You can always review and publish the draft from Agent Studio instead.

5. Test an agent (optional)

The CLI's test command still uses an optional personal tsf_* token:

trueserve test <agent-id> --voice
trueserve test <agent-id>
trueserve test <agent-id> --published

Re-run trueserve configure and supply the optional token if you skipped it. The token needs agent management access. Voice testing opens a local microphone page; the token remains in the CLI process and is not placed in the browser.

Notes