MCP test automation

MCP test automation: let AI agents and your team drive one test platform

Claude Code or any MCP client creates, runs and analyzes your web, mobile and API tests, in the repository your team reviews.

See the 152 MCP toolsOff by default · in every plan

Figure 1. Illustration built from real Cerberus screens.

What is MCP test automation?

Definition

MCP test automation means letting an AI agent create, run and analyze automated tests through tools it calls over the Model Context Protocol (MCP), an open standard. The Cerberus Testing MCP server connects the agent to your test platform.

  • Tests that land in your repository

    What the agent writes becomes a test case your team can open, not a chat.

  • Verdicts from actual runs

    Verdicts and evidence come from an actual execution, never from a model’s guess.

  • An open standard, no lock-in

    Claude Code or any MCP client connects; Cerberus Testing has been open source since 2013.

  1. Anthropic introduces the Model Context Protocol as an open standard. Source (opens in a new tab)
  2. MCP joins the Agentic AI Foundation, under the Linux Foundation. Source (opens in a new tab)

Two kinds of MCP servers for testing

An MCP server for testing either drives a browser or works on the tests themselves. Cerberus Testing’s belongs to the second family.

Browser MCP servers: Playwright MCP, Chrome DevTools MCP

They give an agent a browser.

The agent can
Open a page, read its accessibility tree, click, type, take screenshots.
Playwright MCP tools
  • browser_navigate
  • browser_snapshot
  • browser_click
  • browser_type
What it produces
Actions in a live browser, and Playwright code for them (--codegen).
Best for
Exploring a page, reproducing a bug.

Test-platform MCP server: Cerberus Testing

It gives an agent your test platform.

The agent can
Create, run and analyze tests, campaigns, schedules, robots and data: 152 tools.
Cerberus Testing tools
  • cerberus_testcase_create
  • cerberus_campaign_execution_create
  • cerberus_tag_wait
  • cerberus_testcase_execution_get
What it produces
Test cases, runs and results your whole team opens in the web interface.
Best for
A regression suite the whole team runs, across web, mobile and APIs.
Figure 2. Not rivals: one MCP client, Claude Code for example, can connect to both.

Playwright MCP gives an agent a browser. Cerberus Testing gives it your whole test platform.

The 152 tools of the Cerberus Testing MCP server

From test folders to robots: 152 tools in 17 areas, each with a single job.

152MCP tools in 17 areas

  • 66 read
  • 28 create
  • 2 launch a test or a campaign
  • 29 update, move or reorder
  • 24 delete or detach
  • 3 pause, resume or cancel a run

24 tools declare it: 23 deletions and cerberus_tag_cancel. Your MCP client can ask for your approval before running one.

Execution

Resolve where a test can run, launch it, read the result and its evidence.
6 tools
  • cerberus_testcase_execution_targetsThe country, environment and robot combinations that can really run a test case.Read
  • cerberus_testcase_execution_createQueues and triggers a test case on one or more countries and environments; returns a tag.Launch
  • cerberus_campaign_execution_createRuns a whole campaign, with optional overrides; returns a tag.Launch
  • cerberus_testcase_execution_listExecutions and queue entries, by tag or by test case.Read
  • cerberus_testcase_execution_getResult of an execution: status, and where and why it failed.Read
  • cerberus_testcase_execution_file_getAn execution artifact: page source, service response, log, or the screenshot as an image.Read

Run control

Wait for a run by its tag and read the verdict, or pause, resume and cancel it.
5 tools
  • cerberus_tag_waitWaits for a run to finish, up to 55 s per call, then returns counters, CI verdict and failures.Read
  • cerberus_tag_getAggregated result of a tag: CI score, flaky tests.Read
  • cerberus_tag_pausePauses what has not started yet.Run control
  • cerberus_tag_resumeResumes a paused run.Run control
  • cerberus_tag_cancelCancels what has not started; running tests finish normally.Run controldestructiveHint

Test cases

Create, read and search test cases, including the sequence that actually runs.
7 tools
  • cerberus_testcase_listLists the test cases of a folder.Read
  • cerberus_testcase_getDetails of a test case, with its steps if asked.Read
  • cerberus_testcase_scenario_getThe sequence that actually runs, library steps resolved.Read
  • cerberus_testcase_search_contentEvery place a string appears: selector, URL, value, property.Read
  • cerberus_testcase_createCreates a test case in a folder; the ID is assigned for you.Create
  • cerberus_testcase_updateUpdates, activates or deactivates a test case.Update
  • cerberus_testcase_deleteDeletes a test case.DeletedestructiveHint

Steps

Steps of a test case, and reusable library steps.
9 tools
  • cerberus_testcase_step_listSteps of a test case, as stored.Read
  • cerberus_testcase_step_getDetails of one step.Read
  • cerberus_testcase_step_createAdds a step, at a given position if needed.Create
  • cerberus_testcase_step_updateUpdates a step.Update
  • cerberus_testcase_step_deleteDeletes a step with its actions and controls.DeletedestructiveHint
  • cerberus_testcase_step_duplicateCopies a step, actions and controls included.Create
  • cerberus_testcase_step_reorderSets the order in which steps run.Update
  • cerberus_step_library_listReusable library steps of a system.Read
  • cerberus_testcase_step_library_useAdds a step that points to a library step: fix it once, and it’s fixed everywhere.Create

Actions

What each step does: open a URL, click, type, call a service…
7 tools
  • cerberus_testcase_step_action_listActions of a step.Read
  • cerberus_testcase_step_action_getDetails of one action.Read
  • cerberus_testcase_step_action_createAdds an action: openUrl, click, type, callService…Create
  • cerberus_testcase_step_action_updateUpdates an action.Update
  • cerberus_testcase_step_action_deleteDeletes an action.DeletedestructiveHint
  • cerberus_testcase_step_action_moveMoves an action, with its controls, to another step.Update
  • cerberus_testcase_step_action_reorderSets the order in which the actions of a step run.Update

Controls

The assertions that decide whether an action passed.
7 tools
  • cerberus_testcase_step_action_control_listControls (assertions) of an action.Read
  • cerberus_testcase_step_action_control_getDetails of one control.Read
  • cerberus_testcase_step_action_control_createAdds a control, such as verifyElementPresent.Create
  • cerberus_testcase_step_action_control_updateUpdates a control.Update
  • cerberus_testcase_step_action_control_deleteDeletes a control.DeletedestructiveHint
  • cerberus_testcase_step_action_control_moveMoves a control to another action.Update
  • cerberus_testcase_step_action_control_reorderSets the order in which the controls of an action run.Update

Countries and properties

Where a test case can run, and the %property.NAME% values it uses per country.
9 tools
  • cerberus_testcase_country_listCountries a test case can run in.Read
  • cerberus_testcase_country_getChecks one country of a test case.Read
  • cerberus_testcase_country_createAdds a country to a test case.Create
  • cerberus_testcase_country_deleteRemoves a country from a test case.DeletedestructiveHint
  • cerberus_testcase_country_property_listProperties a test case can use, written %property.NAME%.Read
  • cerberus_testcase_country_property_getOne property, with its countries.Read
  • cerberus_testcase_country_property_createCreates a property, for one country or all.Create
  • cerberus_testcase_country_property_updateUpdates a property for a country.Update
  • cerberus_testcase_country_property_deleteDeletes a property for a country.DeletedestructiveHint

Test folders

The folders that organize the test repository.
5 tools
  • cerberus_test_folders_listLists the test folders.Read
  • cerberus_test_folder_getDetails of one test folder.Read
  • cerberus_test_folder_createCreates a test folder.Create
  • cerberus_test_folder_updateUpdates, activates or deactivates a test folder.Update
  • cerberus_test_folder_deleteDeletes a test folder.DeletedestructiveHint

Labels

Test case labels, system labels, and the labels that select a campaign’s tests.
10 tools
  • cerberus_testcase_label_listLabels attached to a test case.Read
  • cerberus_testcase_label_addLabels a test case, which puts it in the campaigns using that label.Create
  • cerberus_testcase_label_removeDetaches labels from a test case.DeletedestructiveHint
  • cerberus_system_label_listLabels defined in a system.Read
  • cerberus_system_label_createDefines a new label in a system.Create
  • cerberus_label_listLabels that select a campaign’s test cases.Read
  • cerberus_label_getOne label attached to a campaign.Read
  • cerberus_label_createAttaches a label to a campaign.Create
  • cerberus_label_updateChanges which label a campaign attachment points to.Update
  • cerberus_label_deleteRemoves a label from a campaign.DeletedestructiveHint

Campaigns

Campaigns, their criteria, cron schedules and notifications.
20 tools
  • cerberus_campaign_listLists campaigns.Read
  • cerberus_campaign_getA campaign’s settings, labels, countries, environments and robots.Read
  • cerberus_campaign_createCreates a campaign: a reusable set of tests and where to run them.Create
  • cerberus_campaign_updateUpdates a campaign.Update
  • cerberus_campaign_deleteDeletes a campaign.DeletedestructiveHint
  • cerberus_campaign_parameter_listCountries, environments, robots and filters of a campaign.Read
  • cerberus_campaign_parameter_getOne campaign criterion.Read
  • cerberus_campaign_parameter_createAdds a criterion: country, environment, robot, status, priority…Create
  • cerberus_campaign_parameter_updateChanges a campaign criterion in place.Update
  • cerberus_campaign_parameter_deleteRemoves a campaign criterion.DeletedestructiveHint
  • cerberus_campaign_scheduler_listCron schedules of a campaign.Read
  • cerberus_campaign_scheduler_getOne schedule: cron, active flag, last run.Read
  • cerberus_campaign_scheduler_createSchedules a campaign with a Quartz cron expression.Create
  • cerberus_campaign_scheduler_updateUpdates a schedule.Update
  • cerberus_campaign_scheduler_deleteRemoves a schedule.DeletedestructiveHint
  • cerberus_eventhook_listNotifications fired when a campaign starts or ends.Read
  • cerberus_eventhook_getOne notification hook.Read
  • cerberus_eventhook_createNotifies Slack, Teams, email, Google Chat or a webhook, optionally only on failure.Create
  • cerberus_eventhook_updateUpdates a notification hook.Update
  • cerberus_eventhook_deleteRemoves a notification hook.DeletedestructiveHint

Applications and objects

Applications under test and their reusable locators.
10 tools
  • cerberus_application_listApplications under test.Read
  • cerberus_application_getDetails of an application.Read
  • cerberus_application_createCreates an application: name, type, system.Create
  • cerberus_application_updateUpdates an application.Update
  • cerberus_application_deleteDeletes an application.DeletedestructiveHint
  • cerberus_application_object_listReusable locators (XPath, CSS…) of an application.Read
  • cerberus_application_object_getOne application object.Read
  • cerberus_application_object_createCreates a locator that actions reference by name.Create
  • cerberus_application_object_updateUpdates a locator once, for every test that uses it.Update
  • cerberus_application_object_deleteDeletes an application object.DeletedestructiveHint

Services

REST, SOAP, Kafka, MongoDB and FTP calls, with headers and content.
15 tools
  • cerberus_appservice_listServices Cerberus can call.Read
  • cerberus_appservice_getA service with its headers and content.Read
  • cerberus_appservice_createCreates a service, straight from a whole curl command.Create
  • cerberus_appservice_updateUpdates a service.Update
  • cerberus_appservice_deleteDeletes a service.DeletedestructiveHint
  • cerberus_appservice_header_listHTTP headers of a service.Read
  • cerberus_appservice_header_getOne header of a service.Read
  • cerberus_appservice_header_createAdds an HTTP header to a service.Create
  • cerberus_appservice_header_updateUpdates a header.Update
  • cerberus_appservice_header_deleteRemoves a header.DeletedestructiveHint
  • cerberus_appservice_content_listContent parameters: form data, URL-encoded.Read
  • cerberus_appservice_content_getOne content parameter.Read
  • cerberus_appservice_content_createAdds a content parameter.Create
  • cerberus_appservice_content_updateUpdates a content parameter.Update
  • cerberus_appservice_content_deleteRemoves a content parameter.DeletedestructiveHint

Data library

Test data from SQL, services, CSV files or internal values.
4 tools
  • cerberus_datalib_listData libraries a getFromDataLib property can use.Read
  • cerberus_datalib_getA data library, its entries and column mapping.Read
  • cerberus_datalib_createCreates a library and its entries in one call: SQL, service, CSV, internal.Create
  • cerberus_datalib_updateCorrects a data library’s definition or mapping.Update

Environments

Country and environment pairs, and how each application is reached.
10 tools
  • cerberus_country_env_param_listCountry and environment pairs enabled for a system.Read
  • cerberus_country_env_param_getOne environment: build, revision, maintenance.Read
  • cerberus_country_env_param_createEnables a country and environment pair (DEV, QA, PROD…).Create
  • cerberus_country_env_param_updateUpdates an environment configuration.Update
  • cerberus_country_env_param_deleteRemoves an environment configuration.DeletedestructiveHint
  • cerberus_country_environment_parameters_listHow each application is reached, per environment.Read
  • cerberus_country_environment_parameters_getEndpoint, context root, login URL and pool size of one application.Read
  • cerberus_country_environment_parameters_createDeclares how an application is reached on an environment.Create
  • cerberus_country_environment_parameters_updateUpdates how an application is reached.Update
  • cerberus_country_environment_parameters_deleteRemoves how an application is reached.DeletedestructiveHint

Robots and executors

Execution profiles, WebDriver capabilities and the hosts that run them.
15 tools
  • cerberus_robot_listExecution profiles: platform, browser, version.Read
  • cerberus_robot_getA robot with its capabilities and executors.Read
  • cerberus_robot_createCreates a robot.Create
  • cerberus_robot_updateUpdates a robot.Update
  • cerberus_robot_deleteDeletes a robot.DeletedestructiveHint
  • cerberus_robot_capability_listWebDriver capabilities of a robot.Read
  • cerberus_robot_capability_getOne capability.Read
  • cerberus_robot_capability_createAdds a WebDriver capability (key and value).Create
  • cerberus_robot_capability_updateUpdates a capability.Update
  • cerberus_robot_capability_deleteRemoves a capability.DeletedestructiveHint
  • cerberus_robot_executor_listHosts able to run a robot.Read
  • cerberus_robot_executor_getOne executor: host, port, proxy.Read
  • cerberus_robot_executor_createAdds a host that runs a robot.Create
  • cerberus_robot_executor_updateUpdates an executor.Update
  • cerberus_robot_executor_deleteRemoves an executor.DeletedestructiveHint

Invariants

Reference lists: countries, environments, browsers, priorities, statuses…
5 tools
  • cerberus_invariant_listValues of a reference list, for one type (countries, environments…).Read
  • cerberus_invariant_getOne reference value.Read
  • cerberus_invariant_createAdds a reference value.Create
  • cerberus_invariant_updateUpdates a reference value.Update
  • cerberus_invariant_deleteDeletes a reference value.Delete

Context and reference

Active workspaces, Cerberus variables, and the catalogs of actions, controls, conditions and property types.
8 tools
  • cerberus_context_system_listYour active workspaces, and the ones you are allowed to activate.Read
  • cerberus_context_system_updateAdds or removes workspaces from your active context.Update
  • cerberus_variable_listVariables Cerberus substitutes in a value, and when each one resolves.Read
  • list_available_actionsActions available for an application type.Read
  • list_available_controlsControls (assertions) available for an application type.Read
  • list_available_conditionsCondition operators that decide whether a step or action runs.Read
  • list_available_property_typesProperty types: text, SQL, JSONPath, OTP…Read
  • get_db_versionVersion of the Cerberus database schema.Read

Names as your MCP client lists them. Source: cerberustesting/cerberus-core (opens in a new tab).

Cerberus Testing and Playwright: complementary

Playwright suits tests versioned with your code; Cerberus Testing suits campaigns shared with QA and business teams.

Playwright and Cerberus Testing compared, criterion by criterion. “Strength” marks where each one stands out.
CriterionPlaywright with Playwright MCP and Test AgentsCerberus Testing with its MCP server
What it isA library and test runner; Playwright MCP drives a browser.A test automation platform with a built‑in MCP server.
Who writes testsDevelopers and SDETs, comfortable with code.The whole team, in a low‑code editor, and AI agents through MCP.Strength
What its MCP server gives an agentA browser: navigate, read the accessibility tree, click, type.Test repository, runs, campaigns, robots and data: 152 tools.Strength
ScopeWeb (Chromium, Firefox, WebKit), API, mobile emulation; Android is experimental.Web (Selenium), native Android and iOS (Appium), desktop (Sikuli), REST, SOAP, Kafka, MongoDB, FTP, SQL, batch.Strength
Where tests liveFiles in your Git repository, with code review and diffs.StrengthA central repository shared by the team; versioning differs from Git.
Reports and debuggingHTML report, Trace Viewer, UI mode.StrengthExecution and campaign reports; screenshots, video, logs, network traffic.
Running at scaleRunner parallelism and sharding, in CI.StrengthCampaigns your team launches or schedules from the web interface: a queue, parallel runs across countries and environments, a CI score.Strength
Built‑in AITest Agents (planner, generator, healer) for test code.AI Assistant (Enterprise extra) that calls the same 152 tools, with your Anthropic key.
LicenseApache-2.0, maintained by Microsoft.GPLv3, open source since 2013.
HostingRuns locally or in CI; no server.Cerberus Testing Cloud, managed for you, from a free plan to Enterprise.
CommunityVery large: about 97,000 GitHub stars.StrengthSmaller; in production at La Redoute and MyTF1.

Playwright facts: playwright.dev (opens in a new tab) and GitHub, September 2026.

Get started: connect Claude Code in 4 steps

A cloud instance, one setting, one key and one command.

  1. Start with Cerberus Testing Cloud

    The free plan gives one user a dedicated instance; the MCP server is included in every plan.

    Free cloud plan

    • 1 user
    • Dedicated instance
    • Nothing to install
  2. Turn on the server, get your API key

    An administrator sets cerberus_mcp_enable to true in Admin › Parameters, then creates your key in Admin › Users.

    Admin › ParametersValue

    cerberus_mcp_enable

    Boolean in order to enable the MCP endpoints.

    falsetrue

    Admin › UsersYour user

    API Key

    Set New KeyCopy
  3. Add the server to Claude Code

    One command, with your instance’s /mcp address and your key in the X-API-KEY header.

    Terminal

    claude mcp add --transport http cerberus https://your-cerberus/mcp --header "X-API-KEY: <your-key>"
    Prefer JSON? Same settings for other MCP clients with remote HTTP support
    {
      "mcpServers": {
        "cerberus": {
          "type": "http",
          "url": "https://your-cerberus/mcp",
          "headers": {
            "X-API-KEY": "<your-key>"
          }
        }
      }
    }
  4. Ask, then read the result

    The agent resolves the targets, runs the test, waits with cerberus_tag_wait and reports the verdict.

    “Run the login test on QA and tell me the verdict.”

What each Cerberus Testing Cloud plan includes

The MCP server comes with every plan; the built‑in AI Assistant is an Enterprise extra.

CapabilityFree plan1 user, dedicated instanceStarter · BusinessManaged SaaSEnterpriseManaged SaaS, plus extras
Cerberus Testing MCP server152 tools at /mcp, off by defaultIncluded in Free planIncluded in Starter · BusinessIncluded in Enterprise
Built‑in AI AssistantIn‑app chat on the same tools, with your Anthropic keyNot included in Free planNot included in Starter · BusinessIncluded in Enterprise
Start for freeCompare plans28‑day free trial on every paid plan.

FAQ

MCP test automation: FAQ

What is MCP in test automation?
MCP test automation means letting an AI agent create, run and analyze tests through the Model Context Protocol, an open standard. The agent calls typed tools and reads results from actual test runs; the Cerberus Testing MCP server exposes a whole test platform.
What can an AI agent do with the Cerberus Testing MCP server?
Through 152 tools, it can write and run tests and campaigns, read verdicts and screenshots, schedule runs and alerts, and manage services, test data, environments and robots, in the same repository your team uses in the web interface. See all 152 tools.
What is the difference between Playwright MCP and the Cerberus Testing MCP server?
Playwright MCP gives an agent a browser to navigate, click and type. The Cerberus Testing MCP server gives it your test platform: repository, runs, campaigns, robots and results. They are complementary rather than alternatives: one MCP client can connect to both.
Which AI clients can connect?
Claude Code, or another MCP client that supports remote Streamable HTTP servers and sends an X-API-KEY header or uses OAuth 2.0 through Keycloak. The MCP Inspector gives you the configuration to copy.
Is the MCP server included in my plan?
Yes. The Cerberus Testing MCP server is included in every Cerberus Testing Cloud plan, the free plan too, and its code is open source (GPLv3) like the rest of Cerberus Testing. It stays off until an administrator turns on cerberus_mcp_enable.
How much does the AI cost?
The MCP server costs nothing extra: it comes with every Cerberus Testing Cloud plan. You pay for the model through your own account: Claude Code runs on your Claude plan or API key, and the built‑in AI Assistant (Enterprise extra) uses your Anthropic key and estimates the cost of each session.
Is it safe to let an AI agent run my tests?
The server is off by default, and every call is authenticated and runs with the Cerberus account of its key or token. MCP tools do not apply Cerberus roles yet, so give your agent a dedicated account. The 24 destructive tools are flagged, so your client can ask before running one; approval happens in the client, not on the server. Calls can be logged.
Where does my test data go?
Your tests, runs and evidence stay in your Cerberus Testing instance. Cerberus Testing calls an AI provider only for its built‑in AI features, and only the one you configure; an external agent such as Claude Code sends what it reads to its own model. Notifications and cloud robots call out only if you set them up.
Do I need to code to automate tests with AI?
No. A Cerberus test is a list of steps, actions and controls in a low‑code editor. An agent can write them; your team reads, edits, runs and schedules them there, without code.
Will AI agents replace QA testers?
No. The agent takes the repetitive work: writing steps, running tests, reading results, fixing a renamed selector everywhere. Your QA team decides what to test, reviews what the agent writes in the low‑code editor and owns the verdict.

Another question?

Our team answers in English, French and Portuguese.

Talk to our team

Get started

Give your agents a test platform, not just a browser

Start MCP test automation with Cerberus Testing Cloud, for free. Your agents get the MCP server; your team gets the web interface.