Coding agents and access

Cursor integration

TesterArmy's hosted MCP server lets Cursor create, run and read tests from its own session in the editor.

START TESTING FOR FREE READ THE SETUP DOCS

[01] Session

Failed runs triaged in Cursor

A Cursor session runs the Keep tests healthy prompt. The agent lists the day's failed runs, reads a transcript and the network log, then hands over the recording link.

Cursor · testerarmy MCP

❯

List the runs that failed in the last day and tell me which failures share a root cause.

MCP (list_runs: status failed, newest first)

failed runs from the last day listed • done

MCP (get_run_transcript: each failed run's transcript)

reasoning and tool calls read in order • done

MCP (get_run_telemetry: console and network logs per run)

console errors and failed requests read • done

MCP (get_run_recording: newest failure's video link)

signed link, expires in 15 minutes • done

Every failed run stops at sign-in. Telemetry shows the login request answering with a server error, so no test reaches its first step and the bug is in the app. The newest failure's recording link is above.

Membership re-checked
Each request is re-checked against current team membership, and losing access to a team stops the tools immediately.
Team choice
Tools ask which team to use when an account has several, and the agent resolves it with list_teams.
Recording link
The run video arrives as a signed link that expires in 15 minutes and never enters the conversation.

[02] Setup

How it works

Setup is one JSON entry in .cursor/mcp.json, per project or global, and the first tool call triggers the browser sign-in, so a No authorization provided reply until then is expected.

Add the server to Cursor

Add a testerarmy server entry pointing at https://tester.army/mcp to .cursor/mcp.json in the project, or to the global ~/.cursor/mcp.json.

Sign in on first use

Cursor asks for authentication the first time a tool is used, then the browser opens for the TesterArmy sign-in over OAuth with no API key.

Ask the agent in Cursor

Ask Cursor in plain language and it saves tests with create_test, starts them with run_test or run_group, and reads failures with get_run_transcript and get_run_telemetry.

Read the setup docsThe docs cover the Cursor mcp.json entry, the tools the agent gets, starting prompts, teams, security and troubleshooting.

See it on your own app

START TESTING FOR FREE

[03] Related

Works with

Claude Code and Codex use the same hosted server, the MCP server page covers Claude Desktop, claude.ai and stdio clients, and the GitHub App tests the pull requests Cursor writes.

Claude CodeClaude Code creates, runs and reads TesterArmy tests from its own session.CodexCodex creates tests, runs saved groups and reads failures over the hosted MCP server.MCP serverClaude, Cursor, Codex and VS Code create tests, start runs and debug failures.GitHubThe GitHub App posts checks and comments on every pull request preview.

[04] FAQ

Questions

Once .cursor/mcp.json points at the hosted server, Cursor saves tests written as plain-language steps, starts a single test or a whole group against a saved environment, polls each run and cancels work in flight.

Cursor prompts for the TesterArmy sign-in the first time a tool is used, and the browser completes it over OAuth with no API key. API keys belong to the ta CLI and the public API.

The tester-army/cli repository ships the testerarmy-cli skill, installed with npx skills add tester-army/cli, and a .cursor-plugin manifest. The skill covers running tests and iterating on failures, and the server is added through mcp.json.

The exploration agent reads the whole pull request description before planning. With the docs' AGENTS.md excerpt added once, every pull request Cursor writes ends with testing instructions naming the entry point, steps and expected behavior.

[05] Start in Cursor

Run the first test from Cursor

Create a TesterArmy account, add the testerarmy entry to .cursor/mcp.json, and complete the OAuth sign-in when Cursor asks on the first tool call.