Instrument with an agent

Talaria's hosted MCP is how a coding agent shares your project. It reads the same docs this site publishes, creates a development key after you agree, and later pulls the error, the trace, and the session when you are fixing a release. Dashboard AI — Investigate on an issue, Explain on an event, Analyse on a replay — is a separate assistant inside the dashboard.

The loop

1. The agent calls docs_get on guides/add-talaria-with-an-agent, then searches the SDK docs for your stack.
2. You connect MCP (OAuth, or a personal token from Settings → Integrations). The agent calls get_connection and tells you which user and organization it sees.
3. You say yes. The agent calls setup_project with confirmed: true. That turns tracing on and returns one development key.
4. You put the key in env or the equivalent define. The SDK does not send an environment — the key decides it. Send a release version or commit SHA.
5. For a browser, React, or Next app, upload source maps for that same release. The walkthrough is Upload JavaScript source maps. The upload key has scope releases:write and stays out of the browser bundle.
6. The agent calls send_test_event, then search_errors or search_events with environment: development.

What the agent can read later

• MCP can list issues, open one error, open the trace, and list sessions in an environment.
• It can list feature flags, turn a flag off, or set a percent rollout. Segments and experiments stay in the dashboard.
• It does not run dashboard AI. Investigate, Explain, and replay analysis stay on the issue, event, and replay pages.

Where to start

Humans and agents use the same playbook: Add Talaria with an agent. JavaScript source maps are uploaded for the release. Cursor setup is on the Cursor page. The product story is Agents.