agent-qa

Name: agent-qa
Author: Vostride AI

Write, run and triage natural-language end-to-end tests for web and mobile apps from a coding agent

Official · Vostride AI TypeScript Local · stdio, Streamable HTTP Testing Browser automation

GitHub npm Website

60 popularity score ★ 855 ↓ 1.7k/wk

Connect your client

Local (once it is running)

Add in one click

Add to Cursor Add to VS Code Add to LM Studio Add to Goose

Opens the app, which asks before it adds the server.

Or add it by command or config

Claude CodeTerminal

claude mcp add --transport http agent-qa http://127.0.0.1:3471/mcp

Cursor~/.cursor/mcp.json

{
  "mcpServers": {
    "agent-qa": {
      "url": "http://127.0.0.1:3471/mcp"
    }
  }
}

VS Code.vscode/mcp.json

{
  "servers": {
    "agent-qa": {
      "type": "http",
      "url": "http://127.0.0.1:3471/mcp"
    }
  }
}

Codex~/.codex/config.toml

[mcp_servers.agent-qa]
url = "http://127.0.0.1:3471/mcp"

Gemini CLITerminal

gemini mcp add --scope user --transport http agent-qa http://127.0.0.1:3471/mcp

Local (npm)

Add in one click

Add to Cursor Add to VS Code Add to LM Studio Add to Goose

Opens the app, which asks before it adds the server.

Or add it by command or config

Claude CodeTerminal

claude mcp add agent-qa -- npx -y agent-qa mcp

Claude Desktopclaude_desktop_config.json

{
  "mcpServers": {
    "agent-qa": {
      "command": "npx",
      "args": [
        "-y",
        "agent-qa",
        "mcp"
      ]
    }
  }
}

Cursor~/.cursor/mcp.json

{
  "mcpServers": {
    "agent-qa": {
      "command": "npx",
      "args": [
        "-y",
        "agent-qa",
        "mcp"
      ]
    }
  }
}

VS Code.vscode/mcp.json

{
  "servers": {
    "agent-qa": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "agent-qa",
        "mcp"
      ]
    }
  }
}

Codex~/.codex/config.toml

[mcp_servers.agent-qa]
command = "npx"
args = ["-y", "agent-qa", "mcp"]

Gemini CLITerminal

gemini mcp add --scope user agent-qa npx -- -y agent-qa mcp

Settings 4

`AGENT_QA_MCP_PORT`	Optional: Port for the dashboard's HTTP MCP endpoint, overriding services.mcp.port (3471 by default).
`AGENT_QA_DASHBOARD_PORT`	Optional: Overrides services.dashboard.port, the dashboard port the run tools talk to.
`AGENT_QA_LOG_LEVEL`	Optional: Overrides services.logging.level, e.g. debug.
`AGENT_QA_HEADLESS`	Optional: Set to false to show the browser during runs (use.browser.headless).

What it does

agent-qa is a test harness from Vostride: tests are YAML files written in plain English, and an LLM you configure carries them out against a real browser, Android device or iOS simulator, repairing failed steps during the run and keeping file-based memory and an action cache between runs. The `agent-qa mcp` subcommand of its CLI starts a local MCP server so a coding agent can work with that harness directly.

Through the server an agent can inspect the active config (keys, tokens, secrets and passwords come back masked), look up the schemas for config, test, suite and hook files, generate and check canonical IDs, and validate definitions. With a running agent-qa dashboard it can also list, create, update and delete tests, suites and hooks, run a hook, queue test or suite runs, read run steps, logs, execution logs and artifacts, cancel a run, and get a rule-based classification of why a run failed. The source registers 33 tools, plus the `agent_qa_authoring_context` prompt and schema resources at `agent-qa://schema/{schema}` (packages/mcp/src/agent-qa-server.ts).

There are two ways to connect: stdio via `agent-qa mcp`, or the Streamable HTTP endpoint the dashboard serves, by default at http://127.0.0.1:3471/mcp. Over stdio, every authoring, run, artifact and triage tool needs a `dashboardUrl` argument; over the dashboard endpoint that context is filled in for you. Vostride also runs a separate, read-only documentation server at https://vostride.com/mcp with two tools for searching and reading the agent-qa docs; it cannot run tests.

What it needs

Runtime	Node.js 24+

Node.js 24 or later (the package's engines field) or Bun, with agent-qa installed in the project (npm install -D agent-qa) and initialised with `npx agent-qa init`. Access to an LLM, preferably multimodal: an OpenAI- or Anthropic-compatible endpoint, Gemini, a local model, or Codex/Claude Code subscriptions via the extra @vostride/agent-qa-subscription-auth package. Browsers via `agent-qa install-browsers`; for mobile, Appium, the mobile drivers and Android SDK or Xcode. Docker for hooks. Most tools need the dashboard running (`agent-qa dashboard`).

Limits

Over stdio only the six discovery, config, schema and ID tools work on their own; the rest fail with "dashboardUrl is required" unless the dashboard is running and its URL is passed. `agent_qa_classify_failure` uses fixed heuristics over run detail, logs and artifacts, not an LLM call, and falls back to `unknown_failure` at 0.3 confidence. Test runs spend tokens on the LLM you configure. The hosted docs server at vostride.com/mcp cannot run tests or read your repository. Version 0.1.21 is an early release.

Who it suits

Teams already using, or ready to adopt, agent-qa for end-to-end tests who want a coding agent to write tests, start runs and read the evidence without leaving the editor. It is a local, read-write server tied to one harness and its YAML format, and it needs an LLM plus browser or mobile tooling set up first. Prefer the HTTP endpoint when the dashboard can stay running. For quick browser driving with no test files, memory or dashboard, a general browser automation server will be less to set up.

Tools 33

agent_qa_discover: Report the server's tools, resources, prompts, endpoint and config paths
agent_qa_get_config: Read the active config, secrets masked, in full or as a summary
agent_qa_schema_reference: Return the schema reference for config, tests, suites, hooks or IDs
agent_qa_validate_definition: Check a config, test, suite or hooks object against the schemas
agent_qa_generate_id: Create a canonical ID for a test, suite, hook, run or observation
agent_qa_validate_id: Check an ID against the expected ID format
agent_qa_list_tests: List the tests in the workspace
agent_qa_read_test: Read one test definition
agent_qa_validate_test: Validate a test definition
agent_qa_create_test: Create a new test file
agent_qa_update_test: Change an existing test
agent_qa_delete_test: Delete a test
agent_qa_list_suites: List the suites in the workspace
agent_qa_read_suite: Read one suite definition
agent_qa_validate_suite: Validate a suite definition
agent_qa_create_suite: Create a new suite
agent_qa_update_suite: Change an existing suite
agent_qa_delete_suite: Delete a suite
agent_qa_list_hooks: List the hooks in the workspace
agent_qa_read_hook: Read one hook
agent_qa_create_hook: Create a new hook
agent_qa_update_hook: Change an existing hook
agent_qa_delete_hook: Delete a hook
agent_qa_run_hook: Run a hook in its Docker sandbox
agent_qa_enqueue_test_run: Queue a run of one test on the dashboard
agent_qa_enqueue_suite_run: Queue a run of a whole suite
agent_qa_get_run: Get a run's details, steps, attempts and suite context
agent_qa_get_run_steps: Get the steps of a run
agent_qa_get_run_logs: Get a run's log rows, filterable by step, level and source
agent_qa_get_run_execution_logs: Get a run's structured execution logs
agent_qa_get_run_artifact: Get a run's sanitised artifacts
agent_qa_cancel_run: Cancel a pending or running run
agent_qa_classify_failure: Sort a failed run into a failure category with supporting evidence

Popularity

60 popularity score, on our list (the bar is 36)

★ 855 ↓ 1.7k/wk

GitHub stars, whole repository	855	25.9 points
Stars in 30 days	+0	0 points
npm downloads, last week	1,710	25.9 points
Official server	Yes	8 points

It lives in the folder packages/cli of vostride/agent-qa, which holds more than this server. Its stars are the whole repository's, so in the score they never count for more than its own downloads.

Measured 2026-11-07. How the score works.

Where it is published

Code	vostride/agent-qa/packages/cli ★ 855, the whole repository
npm	agent-qa 0.3.3, 2026-11-07

What it can reach

Runs	On your machine
Reaches	Drives a browser, Files on your machine, Your code, The internet, Runs shell commands
Changes things	Yes: it can write, send, create or delete
Sign-in	Other
Licence	FSL-1.1-ALv2 (npm)
Last code change	2026-11-07

Runs on your machine from the project's agent-qa install. Tools can create, change and delete test, suite and hook files in the workspace, queue and cancel runs, and read run logs and artifacts. Runs drive real browsers or devices against the app under test and send screenshots and page state to the LLM provider you configure. `agent_qa_run_hook` executes hook scripts (Node, Bun, Python or Bash) inside Docker containers, and workspace secrets are passed to hooks as environment variables. Model keys live in ~/.agent-qa/auth.json; the config tool masks key, token, secret and password values. The dashboard endpoint binds to 127.0.0.1 by default and the docs advise keeping it on loopback and reviewing client auto-approval before allowing the write tools.

Over time

GitHub stars, whole repository 855 −47 (−5 %)

924855

8 Oct 20267 Nov 2026

npm downloads a week 1.7k +1.7k (+17000 %)

1.7k1

9 Mar 20267 Nov 2026

Before the site measured it, read from npm's daily download counts.

Popularity score 60 −0

6551

8 Oct 20267 Nov 2026

What is rising across the directory

Sources

Page checked 2026-10-14.

Badge for your README

Shows this server's rank in its category, and stars for a server that runs on your machine. It updates every hour.

MarkdownREADME.md

[![agent-qa on EveryMCPServer](https://everymcpserver.com/badge/agent-qa.svg)](https://everymcpserver.com/servers/agent-qa)

Connect your client

Add in one click

Or add it by command or config

Add in one click

Or add it by command or config

Settings 4

What it does

What it needs

Limits

Who it suits

Tools 33

Similar servers

agent-device

Reticle

agent-qa documentation MCP

autocontext

Vibium

Watch Skill