Skip to main content
Glama
kadeep-ai

kadeep mcp

Official
by kadeep-ai

KaDeep Studios is an autonomous release-readiness QA agent. You give it your product and the user journeys that make revenue, and it runs them on web, mobile and desktop. What comes back is a go or no-go, not a test report.

KaDeep CLI brings KaDeep Studios to the command line, as two open-source npm packages:

Package

Command

What it does

kadeep

kadeep

The KaDeep Studios command line and MCP server. Type kadeep for an interactive home: talk to the KaDeep agent, / for commands, search, open any run without copying an id. Browse projects, suites and test cases; run them and read results; gate localization; set up a repository; and let Cursor, Claude Code and other AI coding agents use KaDeep as tools. --json on every command.

releasegate

releasegate

KaDeep Release Gate, a go / no-go quality gate for CI/CD. It reads releasegate.yml, runs the KaDeep Studios checks the policy requires for the current commit, writes a report and passes or fails the build. It has a non-blocking shadow mode for first-time setup.

Contents

Related MCP server: squish-mcp

Quick start

You need Node.js 20+ and a KaDeep Studios account (get started for free).

npx kadeep                       # the interactive home: sign in, pick a project, then ask the agent or type /
npx kadeep run --suite smoke     # or one command at a time: run a suite and wait for the verdict
npx kadeep init                  # add a releasegate policy + GitHub Actions workflow to this repo

After kadeep init, every pull request runs npx releasegate and gets a go / no-go:

KaDeep Release Gate 0.3.0 · Engineering release confidence.
project p_9a118f0499704752be72 · commit 8c3f2d1 on feature/checkout PR #214 · mode enforce
▸ Smoke
✓ Smoke: 12/12 passed (2m 41s)
▸ Checkout
✗ Checkout: 3/4 passed (1m 05s)
    ✗ Pay by card [DEFECT] · Expected "Order placed" to be visible

NO-GO · 1/2 required checks passed · 3m 47s

What it looks like

A terminal UI built on the dot, like the KaDeep Studios logo: every test is a dot, and the release verdict is drawn in dots. It has no dependencies, and output stays plain in CI.

Features

You want to…

Command

Drive everything from one place in the terminal, like Claude Code

kadeep (the interactive home)

Ask the KaDeep agent why something failed, or have it act

Type the question in the home, or kadeep ask "…"

Run AI end-to-end tests from the terminal

kadeep run --suite smoke

Run specific test cases on another browser or screen size

kadeep run --test checkout-works --browser firefox --viewport mobile

See why a test failed, step by step

/runs in the home, then pick the run · kadeep runs show <runId>

Browse projects, suites, test cases and defects

kadeep projects · kadeep suites · kadeep tests · kadeep issues

Block a release when required checks fail

npx releasegate in CI

Try the gate first without blocking anyone

mode: shadow in releasegate.yml

Set up a repository (policy, CI workflow, agent config) in one step

kadeep init

Let Cursor or Claude Code run and read your tests

kadeep mcp (MCP server)

Check localization quality in CI

kadeep loc validate · a localization check in releasegate.yml

Script anything, parse the output

--json on every command, stable exit codes

Publish test results to CI dashboards

kadeep run --junit results.xml · releasegate writes junit.xml

Watch runs live in the terminal, plain in CI

Rich view in a terminal; NO_COLOR, KADEEP_UI=plain

From commit to go / no-go

flowchart LR
    A["Pull request<br/>or push"] --> B["CI runs<br/>npx releasegate"]
    B --> C["Reads<br/>releasegate.yml"]
    C --> D["KaDeep Studios runs the checks<br/>suites · test cases · localization"]
    D --> E{"Verdict"}
    E -->|GO| F["Merge and ship"]
    E -->|NO-GO| G["Build fails<br/>report + PR annotations"]
    E -.->|shadow mode| H["Reported only<br/>never blocks"]
  • GO: every required check passed.

  • NO-GO: a required check failed or errored. An empty suite is never a pass.

  • Shadow mode: the verdict is reported but never fails the build, so you can watch the gate for a while before letting it block.

The report lands in .releasegate/ (report.json, report.md, junit.xml). On GitHub it also goes into the job summary and annotates failing checks.

How it works

flowchart TB
    dev["Developer terminal"] --> kadeep
    agent["AI coding agents<br/>Cursor · Claude Code"] -->|MCP over stdio| kadeep
    ci["CI/CD pipeline<br/>GitHub Actions · GitLab CI · CircleCI"] --> rg
    rg["releasegate<br/>policy · verdict · reports"] -->|npm dependency| kadeep
    kadeep["kadeep<br/>CLI · MCP server · client library"] -->|HTTPS| api["KaDeep Studios<br/>api.kadeep.ai"]
    api --> surfaces["Web · iOS · Android · Desktop"]
  • kadeep is the foundation. It contains the CLI, the MCP server and the client library: sign-in, runs, localization and CI detection. It has no runtime dependencies.

  • releasegate is built on kadeep. It adds the policy file, the go / no-go decision, shadow mode and the reports. It depends on kadeep, never the reverse.

  • Only HTTPS calls to the KaDeep Studios API. Tests run in KaDeep Studios, on the model key your team configured there. The CLI itself never calls an AI model and collects no telemetry.

Runs are jobs on KaDeep's queue. The CLI starts them and polls, so no request stays open while a suite runs. A long suite survives proxies with idle timeouts, and a CI network blip costs one retried poll:

sequenceDiagram
    autonumber
    participant CI as releasegate (CI job)
    participant API as KaDeep Studios API
    participant Q as KaDeep workers
    CI->>API: POST /api/ci/:project/run  { async: true }
    API-->>CI: 202 · job id
    API->>Q: queue the suite
    loop every few seconds
        CI->>API: GET /api/ci/:project/jobs/:id
        API-->>CI: running · 3/12
    end
    Q-->>API: suite run finished
    CI->>API: GET /api/ci/:project/jobs/:id
    API-->>CI: done · results per test
    Note over CI: verdict → report.json · report.md · junit.xml → exit code

Use cases

Run end-to-end tests from the terminal

kadeep use "Acme web"                        # default project
kadeep run --suite smoke                     # ✓ / ✗ per test, exit 1 if any fail
kadeep runs show r_bf6382d90aab              # every step, the verdict and why
kadeep run --suite regression --no-wait      # queue it, come back later
kadeep jobs show job_8ad6a6bb92 --wait

Add a release gate to GitHub Actions

kadeep init generates .github/workflows/releasegate.yml. Its core:

name: KaDeep Release Gate
on:
  pull_request:
  push:
    branches: [main]
jobs:
  releasegate:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: 22
      - run: npx -y releasegate@^0.1
        env:
          KADEEP_CI_TOKEN: ${{ secrets.KADEEP_CI_TOKEN }}

The policy file sits next to it:

# releasegate.yml
version: 1
project: p_9a118f0499704752be72
mode: shadow            # switch to enforce once you trust the verdicts
checks:
  - suite: smoke
  - tests: [checkout-works]
    viewport: mobile
  - localization:
      locales: [hi-IN, ar-AE]

GitLab CI, CircleCI, Bitbucket Pipelines, Buildkite, Azure Pipelines and Jenkins work the same way. See releasegate: CI setup.

Let AI coding agents run your QA (MCP)

claude mcp add kadeep -- npx -y kadeep mcp        # Claude Code
{ "mcpServers": { "kadeep": { "command": "npx", "args": ["-y", "kadeep", "mcp"] } } }

That second snippet goes in Cursor's .cursor/mcp.json. Your agent can then answer "which smoke tests fail and why?" or "run the checkout tests" by itself. It gets 14 tools: projects, suites, test cases, runs, jobs, defects and the localization gate.

Gate localization in CI

kadeep loc push --file locales/en.json --translate     # new source strings
kadeep loc validate --locale hi-IN --locale ar-AE      # exit 1 with the reasons
kadeep loc pull --asset en.json --locale hi-IN         # approved translations only

Which package do I need?

If you…

Use

work in a terminal or write scripts

kadeep

want Cursor, Claude Code or another MCP client to use KaDeep

kadeep mcp

want CI to pass or fail a build on KaDeep results, with a policy and reports

releasegate (it installs kadeep for you)

just want to run one suite in CI and fail on red

kadeep run --project <id> --suite smoke

Terminology

Term

Meaning

KaDeep Studios

The QA platform and app where your projects, test cases and runs live

KaDeep CLI

This repository's command line, the kadeep package and command

KaDeep Release Gate

The CI gate, the releasegate package and command. Its policy file is releasegate.yml

Test case

One user journey KaDeep runs, e.g. checkout works. kadeep tests lists them; the API calls them flows

Suite

A named group of test cases, e.g. smoke

Run / suite run

One execution of a test case / of a suite, with a verdict and evidence

Job

A queued or running execution. The CLI follows jobs until they finish

CI token

A project's token for pipelines (KADEEP_CI_TOKEN), created with kadeep ci-token create

GO / NO-GO

The gate's verdict. Shadow mode reports it; enforce mode fails the build on NO-GO

FAQ

A release gate is an automated check in your CI/CD pipeline that decides whether a change may ship. releasegate makes that decision from KaDeep Studios results for the exact commit under test: suites, test cases and localization. It returns GO or NO-GO, with a report of why.

Yes. kadeep and releasegate are open source under Apache-2.0. They connect to KaDeep Studios, where your tests run, and you can get started for free.

Any pipeline that can run Node.js 20. The commit, branch, PR number and run link are read automatically on GitHub Actions, GitLab CI, CircleCI, Bitbucket Pipelines, Buildkite, Azure Pipelines and Jenkins. Anywhere else, the gate falls back to git.

Yes. kadeep mcp is a Model Context Protocol server over stdio. Any MCP client can list your tests, run them, wait for the verdict and read failures step by step.

It only makes HTTPS calls to the KaDeep API you point it at (https://api.kadeep.ai by default), with no telemetry. Your session is stored in ~/.config/kadeep/config.json, readable only by you. In CI, a per-project CI token is used instead of a personal login.

Yes. Set mode: shadow in releasegate.yml (kadeep init starts there). The gate runs every check, writes the report and says "would have blocked", but always exits 0. Switch to mode: enforce when its verdicts match your judgement.

Development

git clone https://github.com/kadeep-ai/kadeep-cli.git && cd kadeep-cli
npm install
npm run ci            # typecheck (JSDoc + checkJs), tests on the local Node, tarball check
packages/
  kadeep/         CLI, MCP server and client library (no runtime dependencies)
  releasegate/    KaDeep Release Gate (depends on kadeep and yaml)
scripts/
  check-packs.mjs guards what npm publishes
  • Sources: plain ES modules with JSDoc types. There is no build step, so what is in packages/*/src is what ships.

  • CI: runs the tests on Node.js 20, 22 and 24.

  • Tarball check: scripts/check-packs.mjs fails a publish if anything other than bin/, src/, README.md and LICENSE would ship, or if a file contains a secret or a source map.

  • Releases: both packages share one version. Publish kadeep first, then releasegate.

Found a bug or have an idea? Open an issue.

License

Apache-2.0 © Kadeep Technologies. KaDeep Studios is a product of Kadeep Technologies (kadeep.ai).

Related MCP Connectors

  • Direct access to Cypress tests results and accessibility reports in your AI workflow.

  • Code intelligence platform for AI agents. 20 tools for architecture, security & impact analysis.

  • Run, debug, and triage tests from your IDE using natural language, no dashboard switching, no manual data transfers. The TestMu AI (formerly LambdaTest) MCP Server is a single remote server exposing four tool suites: HyperExecute — analyze your project, generate YAML configs and test runner commands, then monitor jobs and sessions. Automation — pull a TestID's details plus command, network, and console logs into one chat for instant root-cause analysis. Includes mobile app upload. SmartUI — explain pixel, layout, DOM, and perceptual changes in a visual regression run, with context-aware React/HTML/CSS fixes. Accessibility — audit any public URL or a local React app against WCAG and get ready-to-apply remediation steps. Connects over https://mcp.lambdatest.com/mcp using OAuth 2.1 — no API keys in your config. One-click install in Cursor; works with Claude, GitHub Copilot, Cline, and any MCP client. Tests execute on the TestMu AI cloud: 3,000+ browsers and 10,000+ real devices.

  • Develop, manage, and debug Railway projects, services, and deployments from within agents.

Related MCP Servers

  • A
    license
    Not graded
    quality
    A
    maintenance
    Autonomous spec-to-product coding-agent CLI. Its MCP server exposes 34 tools over stdio: project state and task-queue ops, memory retrieve/store, code search, quality and verification reports, repo hotspots/co-changes, and structured findings/learnings.
    7,848 npm
    1,077
    Business Source 1.1
  • A
    license
    Not graded
    quality
    D
    maintenance
    Enables AI models (Claude, ChatGPT, GitHub Copilot) to run and analyze local tests, rerun failures, and orchestrate QA workflows using existing UI and API test frameworks.
    8 npm
    MIT