Launchable MCP server: there isn't one

Test selection is exactly the kind of prediction an agent would want to query, and neither Launchable nor CloudBees exposes it.

Last verified 1 September 2026 · from The launchableinc and cloudbees GitHub organizations and the official MCP registry

This page is one of 90 in a directory of monitoring and developer tools. Each page checks what an AI coding agent can get out of the tool through its MCP server, the connector that lets the agent query the tool directly.

What is the Launchable MCP server, and is it usable?

Launchable, now sold as Smart Tests, uses machine learning over test history. It predicts which tests are most likely to fail for a given change, so a pipeline can run a subset instead of everything.

It works as a CLI inside the pipeline: the CLI records builds and test sessions, then returns a subset to run. That is the whole integration surface.

The product lives in the pipeline rather than in a queryable database. The missing server costs you nothing at run time. It costs you when someone asks why a test was skipped.

Every one of the 90 vendors in this directory gets the same four checks. Launchable passes 0 of them.

Launchable works across code, tests and CI, and publishes nothing for any stage.

Launchable does not let an agent ask why a test was skipped. Launchable made that prediction and keeps the model that produced it. Its only interface is a CLI inside the pipeline, and a CLI cannot be queried.

What can an AI agent do with Launchable?

Launchable's acquisition can suggest MCP coverage that the product does not actually have.

Do you need the MCP server at all?

No, because Launchable publishes only a CLI. The CLI is the working route and the only one. It records builds and test sessions and returns the subset to run, all inside the pipeline, so the integration happens where the tests execute rather than where an agent asks questions.

For test results an agent can query afterwards, several vendors do publish servers. ReportPortal exposes 18 tools over launches, test items and logs. BuildPulse covers flaky tests in 5 tools, for under a thousand tokens. Allure TestOps documents 13 tools over test cases and results.

For flakiness specifically, which overlaps with what Launchable predicts, BuildPulse and Trunk both expose it, and Trunk labels three of its four tools experimental.

For the pipeline that runs the subset, Jenkins, CircleCI and Harness all publish maintained servers that explain why a build did what it did.

Keep Launchable in the pipeline through its CLI. Connect a test-results server when an agent needs to ask what happened.

Launchable MCP server setup

Nothing here installs a Launchable server, because there is none, and CloudBees publishes no Smart Tests tools either. This is the working route.

Launchable runs in the pipeline, not in a conversation. Its CLI selects the subset of tests to run for a change, and that stays where it is.

What an agent can have is the outcome: which tests ran, which failed, and how flaky they have been. Several vendors here publish servers for exactly that.

claude mcp add-json "reportportal" '{"command":
  "reportportal-mcp-server",
  "args":["--rp-host","https://rp.internal",
          "--project","my-project","--analytics-off"],
  "env":{"RP_API_TOKEN":"<your-token>"}}'

claude mcp add-json "buildpulse" '{"command":"npx",
  "args":["-y","@buildpulse/mcp"],
  "env":{"BUILDPULSE_TOKEN":"<your-token>"}}'

Every other client

Each block below is the configuration for one client, with the file path and the key that client expects.

Launchable CLI  ->  which tests to run   (not exposed)
Results server  ->  what happened        (reachable)

Read-only access, permissions and security

• No Launchable credential is created

  With no server, nothing here asks for a Launchable or CloudBees token.

• Test results carry less sensitive data than most connectors do

  Names, durations and pass rates carry little that is confidential, which makes this a comparatively safe connector to attach.

• Failure output is the exception

  Stack traces and assertion messages can contain fixture data, hostnames and occasionally secrets, and those reach the model with the result.

• --analytics-off is worth setting

  ReportPortal documents it, and turning usage reporting off is the right default when an agent is driving the queries.

• Scope the token to one project

  Both servers read whatever their token allows, so a project-scoped credential keeps an agent inside the suite you meant.

Troubleshooting

You cannot find a Launchable MCP server

There is not one. The launchableinc organization holds 12 public repositories and none is MCP-related, as of 26 August 2026.

CloudBees acquired Launchable, so its server should cover Smart Tests

It does not. A search of the cloudbees organization for MCP repositories returns zero results, and the hosted CloudBees MCP server carries no Smart Tests tools.

`launchableinc.com` redirects and you cannot find the documentation

The domain redirects to CloudBees, which is where the product went. That redirect is a sign of the acquisition rather than of a connector living elsewhere.

You want the test-selection decision inside an agent conversation

No tool returns it. Launchable's value is choosing which subset of a suite to run for a given change, and that prediction is not exposed over MCP.

Launchable MCP server: Reference

What engineers report

The searches cover both the original vendor and the acquirer, because the second is the likelier source of a false positive.

The launchableinc and cloudbees GitHub organizations were both searched for MCP repositories on 1 September 2026, returning zero in each case. The official MCP registry has no Launchable entry.

Should you connect an agent to Launchable?

No. There is nothing to connect. Best for the CLI in the pipeline, plus a test-results connector for questions after the run.

Strongest fit

Teams running Smart Tests in a pipeline who do not need to query it afterwards.

Main advantage

Launchable's advantage stays inside the pipeline: the CLI returns the subset to run without any connector at all.

Main weakness

The test-selection decision, which is the product's whole output, cannot be queried.

Operational risk

Reading test results and mistaking them for Launchable's selection. The prediction is not in them.

Frequently asked questions