Skip to content
99tests Blog
Visit 99tests
How 99tests works

Order Real-Device App Testing From Your Coding Agent: The 99tests MCP Server

Add the 99tests MCP server to Claude Code, Cursor or VS Code and order a human-run test on real phones and browsers. Verified bugs come back as fix prompts your agent can apply.

3 min read
Hand-drawn sketch of a laptop with a terminal prompt, a phone, a bug and a tick
In this article
  1. What MCP is, in one paragraph
  2. Install in one line
  3. The five tools
  4. How a test cycle runs
  5. What you can ask your agent
  6. Packages and prices
  7. Why a human test belongs in an AI workflow
  8. Where to find it

Coding agents like Claude Code and Cursor can now build a feature, write its tests and run them in minutes. What they can't do is pick up a budget Android phone on a weak Jio signal, switch to the SMS app to read an OTP, and notice that the app reopened on the wrong screen.

The 99tests MCP server closes that gap. Your coding agent can order a human-run test of your app or website on real phones, laptops and browsers, then pull back the verified bugs as fix prompts it can act on. There's no account, no sales call and a fixed price.

What MCP is, in one paragraph

The Model Context Protocol (MCP) is an open standard that lets AI assistants call outside tools. Add an MCP server to your coding agent and the agent gets new abilities, such as reading a database, opening a ticket or, here, ordering a test.

Install in one line

The server is remote and hosted, so there's nothing to install or run. The endpoint is:

https://99tests.com/api/mcp

Claude Code:

claude mcp add --transport http 99tests https://99tests.com/api/mcp

Cursor or VS Code (mcp.json):

{
  "mcpServers": {
    "99tests": {
      "url": "https://99tests.com/api/mcp"
    }
  }
}

Any other client that supports remote MCP servers over Streamable HTTP can use the same URL.

The five tools

ToolWhat it does
list_packagesLists test packages with sessions, devices, turnaround and prices
get_sample_reportShows a sample report, so the agent (and you) can see the format first
create_test_jobCreates a test job for a build (APK, TestFlight link) or URL and returns a checkout link
get_job_statusReturns where the job is: awaiting payment, testing, verifying or delivered
get_reportReturns the verified findings for a delivered job, each with a fix prompt

How a test cycle runs

  1. You ask your agent, for example: "List the 99tests packages and create a Standard job for this build."
  2. The agent calls create_test_job and gives you a checkout link.
  3. You pay in your browser. Payment never goes through the assistant, and the agent never sees card details.
  4. Real testers run the sessions on devices they own: budget Androids, iPhones, Chrome and Safari, on real networks across India.
  5. A reviewer verifies every finding before it goes into the report.
  6. Your agent calls get_report. Each finding comes with the device, OS, steps to reproduce, expected and actual results, screenshots, and a fix prompt it can apply.

What you can ask your agent

  • "Show me a sample 99tests report before we buy anything."
  • "Create a Starter job for the staging URL and focus on sign-up and checkout."
  • "Check the status of our 99tests job."
  • "Pull the 99tests report and fix the P1 bugs, one commit each."

Packages and prices

Prices are fixed and the same through the MCP server as on the website:

PackagePriceSessionsDevicesReport
Starter₹9,999 + GST / $1295 × 45 min23 business days
Standard₹18,999 + GST / $24910 × 45 min43 business days
Deep₹34,999 + GST / $44920 × 90 min65 business days

Looking up packages and the sample report is free.

Not included: test automation, security or penetration testing, performance or load testing, and certification.

Why a human test belongs in an AI workflow

AI-written tests tend to check what the code already does. They're weak at the things users actually hit: interruptions, late OTP SMS, OEM battery savers killing the app, network drops in the middle of a payment, and screens that technically work but confuse people. We wrote about this in how to test AI-generated test cases before release.

The MCP server puts that human check inside the same loop: the agent builds, a person tests on a real phone, and the agent fixes what was found.

Where to find it

The server is listed on the official MCP Registry as com.99tests/mcp. Full docs are on the 99tests MCP server page, and you can see the report format on the sample report.

99tests has run crowdsourced testing since 2010, with 20,000 registered testers across India. Questions about the MCP server go to hello@99tests.com.

Keep reading