Skip to main content

TestRelic MCP 3.1: Ask AI, marketplace, apps, artifacts, and cloud sessions

· 3 min read
TestRelic Team
TestRelic Team
TestRelic Maintainers

The TestRelic MCP server started as a way to give AI coding assistants real testing context — projects, coverage, healing, impact analysis. With 3.1, it grows into a full AI-workflow surface: five new capability sets, a one-click Cursor Marketplace install, and a cleaner tool namespace. It's published as @testrelic/mcp v3.1.2.

Five new capability sets​

MCP capabilities are opt-in via --caps (or the TESTRELIC_MCP_CAPS env var); core is always on. On top of the existing coverage, creation, healing, impact, triage, signals, and devtools, 3.1 adds five:

  • ai — the Ask AI surface: tool catalog, a single-turn agent, conversation management, and artifact generators (dashboards, reports, test plans, presentations, navigation paths).
  • marketplace — first-class testing integrations (Jira, GitHub Actions, BrowserStack, LambdaTest, Sentry, Amplitude, Grafana Loki): catalog, connect / OAuth, and a unified invoke.
  • apps — a generic action runner for connected apps (Slack, Notion, Linear, HubSpot, Google Calendar, and more): list apps, list actions, connect, and a universal execute.
  • artifacts — list, get, export (PNG, PDF), and save-to-file for artifacts the Ask AI agent produces.
  • sessions — cloud test session workspaces (BrowserStack / LambdaTest): search, video, screenshots, logs, and a multi-pane workspace render.

Mix and match them for the agent you're building. The full list is on the Capabilities page.

One-click install in Cursor​

The fastest way to try it is the Cursor Marketplace:

  1. Open Cursor and search Settings → Plugins for testrelic-mcp.
  2. Click Install. Cursor wires up the MCP server automatically.
  3. Open the agent and ask something like "list my TestRelic projects".

The marketplace default boots in mock mode — no account, no token, no network calls — so you can explore every tool safely before connecting anything. When you're ready for real data, create a personal access token (tr_mcp_*) at https://platform.testrelic.ai/settings/mcp-tokens, run npx @testrelic/mcp login, and drop --mock-mode. The Quickstart has the details.

A cleaner tool namespace​

Every tool now uses a tr_* prefix — tr_list_repos, tr_coverage_report, tr_heal_run, tr_analyze_diff, and so on. The old testrelic_* names still work as deprecated aliases, so existing integrations don't break; new work should use tr_*. The full mapping is in the Tools Reference.

We also corrected the default cloud URL to platform.testrelic.ai across the board, so a fresh install points at the right place with no configuration.

Get started​

If you use Cursor, install the plugin and start in mock mode — it's the lowest-friction way to see what the agent can do. For Claude Desktop, VS Code, or any other MCP client, the Overview covers local setup. This release is part of a broader day of launches — see the Python SDK announcement for the rest.

FAQ: Do my existing testrelic_* tool calls still work?

Yes. The testrelic_* names are kept as deprecated aliases for backward compatibility. New tools and docs use the tr_* prefix.

FAQ: Does the Cursor install touch my real data?

No. The marketplace default runs in mock mode with no token and no network calls. You opt into real data by adding a tr_mcp_* token and removing --mock-mode.