Launchable MCP server: use the Smart Tests CLI instead
Launchable predicts which tests a change needs, and it now sells as CloudBees Smart Tests. Agents read its data through the `smart-tests` CLI rather than over MCP.
Checked 3 October 2026 · from CloudBees Smart Tests documentation, the CloudBees Unify MCP tool reference, the cloudbees-oss GitHub organization and the official MCP registry
One of 91 guides to monitoring and developer tools. Each covers what an AI coding agent can get out of the tool through MCP, the protocol that lets the agent query the tool directly.
Summary
Launchable is now CloudBees Smart Tests, and launchableinc.com redirects to the CloudBees product page. CloudBees has not published an MCP server for Smart Tests as of 3 October 2026.
CloudBees documents a different route. The smart-tests view command prints JSON to standard output, which CloudBees describes as easy to feed into an AI coding agent.
A Smart Tests Agent Skill answers product questions from the documentation. It arrived in cloudbees-oss/agent-skills on 30 March 2026.
How Launchable works with MCP
Smart Tests uses test history to predict which tests a change is likely to break. A pipeline then runs that predicted subset instead of the full suite.
The CLI does the pipeline work. record build, record session, then subset or record tests run in that order on each build.
Four view commands read data back as JSON: flaky-tests, longest-tests, never-failing-tests and test-results. Adding --logs to test-results includes stdout and stderr from failed runs.
smart-tests inspect subset --subset-id <id> --json describes one subset decision. Each row gives a test's rank, whether it was included, and its estimated duration.
The CLI source moved from launchableinc/cli to cloudbees-oss/smart-tests-cli, which had 27 stars on 3 October 2026.
Every vendor here gets the same four checks: official, maintained, distributed and usable. Launchable passes 0 of them.
| Check | Result | Evidence |
|---|---|---|
| Official | Fail | CloudBees documents a CLI and an Agent Skill as the agent route for Smart Tests. It has not published a Smart Tests MCP server as of 3 October 2026. |
| Maintained | Not applicable | Not applicable without a server. The CLI source, cloudbees-oss/smart-tests-cli, was last pushed on 2 October 2026. |
| Distributed | Not applicable | The official MCP registry returned nothing for Launchable or Smart Tests on 3 October 2026. CloudBees' one entry, io.github.cloudbees/unify-mcp-server, serves CloudBees Unify. |
| Usable | Not applicable | Not applicable to MCP. The documented route needs a terminal agent, the smart-tests CLI and an API key in SMART_TESTS_TOKEN. |
| Part of the work | What Launchable has here | Can an agent reach it |
|---|---|---|
| Code, tests & review | Flaky tests, failure history, slow and never-failing tests | Yes, as JSON from smart-tests view, run in a shell |
| CI & release | Predictive test selection inside the pipeline | Partial. inspect subset shows rank and inclusion, not the model's reasons |
A terminal agent can work through flaky and failing tests using Smart Tests data today. It can see how a subset ranked each test, but not why the model ranked it there.
What can an AI agent do with Launchable?
The CloudBees Unify MCP server returns nothing about Smart Tests
CloudBees runs an MCP server for its Unify platform at https://mcp.cloudbees.io/v1/mcp. Its tool reference lists nine toolsets, covering workflows, feature flags, security, policy and reports.
None of those toolsets mentions Smart Tests, subsets or test selection.
Guard: Use the smart-tests CLI for test data. Keep the Unify server for CloudBees workflows and feature flags.
smart-tests view misses a flaky test you know about
view flaky-tests defaults to one week of data and 50 results per week. Its --weeks option goes up to 12, and --limit up to 500.
Guard: Pass --weeks and --test-suite explicitly before reading a short list as a healthy suite.
The agent's shell holds the key the pipeline uses
The CLI reads its API key from SMART_TESTS_TOKEN. The same variable authenticates the record and subset commands that write pipeline data.
Guard: Export the key only in the session used for investigation, and keep pipeline keys in CI secrets.
Do you need the MCP server at all?
When the agent cannot run a shell, a test-results MCP server is the nearest substitute. ReportPortal, BuildPulse and Allure TestOps publish servers that explain what ran and failed, though none predicts a subset.
For flakiness alone, BuildPulse and Trunk publish servers that overlap with view flaky-tests.
For the rest of the CloudBees platform, the Unify MCP server covers workflows, CI controllers, feature flags and security findings.
Use the smart-tests CLI for Smart Tests data. Add a test-results MCP server when the agent has no shell.
Launchable integration setup
CloudBees documents two pieces: an Agent Skill for questions about the product, and CLI output for your own test data.
Smart Tests skill in Claude Code
The plugin comes from CloudBees' own marketplace, which has to be added first. It answers questions about Smart Tests from the product documentation.
/plugin marketplace add cloudbees-oss/agent-skills /plugin install smart-tests@cloudbees
Smart Tests skill in other agents
For agents other than Claude Code, CloudBees documents the skills CLI, run from a terminal.
npx skills add cloudbees-oss/agent-skills
Flaky-test data for an agent
Write four weeks of flaky-test data to a file. Then ask the agent to find each test's source, explain the likely cause and propose a fix.
export SMART_TESTS_TOKEN=<your-api-key> smart-tests view flaky-tests --weeks 4 > flaky-tests.json
What the vendor documents
CloudBees suggests three agent workflows for view output. They are fixing flaky tests, reading one test's failure history, and ranking tests whose weeklyScoreDelta shows rising flakiness.
Should you connect an agent to Launchable?
Yes, for terminal agents that can run the `smart-tests` CLI with an API key. Best for teams already running Smart Tests in CI who want an agent to fix flaky tests.
- Strongest fit
- Yes, for terminal agents that can run the
smart-testsCLI with an API key. - Main advantage
- Flaky-test and failure-history data arrives as JSON that CloudBees shaped for agents.
- Main weakness
- There is no MCP endpoint, so chat clients without a shell cannot reach Smart Tests data.
- Operational risk
- Expecting the CloudBees Unify MCP server to answer test-selection questions it does not cover.
From Oodle
One platform for agent traces and infrastructure
Agent traces usually sit in a different product from the rest of your telemetry, so when a slow database makes an agent slow the symptom is in one tool and the cause is in another. Oodle keeps both in one query surface.
See agent observabilityFrequently asked questions
Does Launchable have an MCP server?
No. As of 3 October 2026 CloudBees has not published one for Smart Tests, the product Launchable became. Agents use JSON from the smart-tests CLI instead.
Does the CloudBees Unify MCP server include Smart Tests?
No. Its tool reference lists nine toolsets for Unify, and none is for Smart Tests or test selection.
Can an agent see why a test was left out of a subset?
Partly. smart-tests inspect subset --json shows each test's rank, inclusion and estimated duration. The model's reasoning is not part of that output.
Where did launchableinc.com go?
It redirects to CloudBees' Smart Tests page. The CLI repository also moved, from launchableinc/cli to cloudbees-oss/smart-tests-cli.
