JAIPilot

io.github.JAIPilot由 GitHub 帳號 JAIPilot 發布v7.2.1更新於 Oct 24, 2026

Ship better Java with your coding agent.

發布方自己的服務：QuickBooks Connector by Meridian

已驗證Streamable HTTP可網頁執行開發者工具AI 與機器學習

概覽

AI 產生的概覽

JAIPilot 是一項 Java 測試服務，讓編碼代理為 Java 儲存庫產生、修復並驗證單元測試，並提升行覆蓋率。

可以做什麼
為選定的類別、路徑或程式碼行產生、修復並穩定單元測試，包含有意義的斷言與失敗路徑覆蓋（generate_tests、fix_tests、stabilize_flaky_tests）。將實測行覆蓋率提升至 90% 以上目標，並提供 100% 覆蓋率工作流程（improve_coverage、coverage_100）。只測試目前變更，或在重構前用特徵測試鎖定現有行為並於之後驗證基準（test_current_changes、lock_behavior、verify_behavior）。檢查專案的程式碼、建置與測試框架，然後執行針對性測試、回歸測試套件和新的覆蓋率報告。
適用情境
適用於編碼代理處理 Java 程式碼庫時，希望比代理單獨產生更快、覆蓋率更高的測試，或在重構前建立行為基準。它是疊加在現有代理之上的選用工具，而非替代品。
執行需求
需要 JAIPilot 帳號（含訂閱與額度），並透過 CLI 登入。需要 Git、JDK 和專案的建置工具；建置需包含 mcp 指令，並以儲存庫路徑啟動。測試與指令在本機執行，服務透過網路存取。
安裝前請注意
選定的原始碼、專案脈絡和指令輸出會傳送到 JAIPilot 託管服務與模型供應商。工作流程會修改儲存庫中的測試檔案，接受變更前請檢視結果和 git diff。使用會消耗 JAIPilot 額度，本機指令預設逾時為一小時。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 JAIPilot，將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Web executable，透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。

交給代理程式安裝

把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式，或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南，先向你展示發布者、端點和登入方式，經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。

閱讀 https://sourceweft.com/mcp/io.github.JAIPilot%2Fjaipilot/install.md，並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前，先告訴我它的發布者、端點和登入方式，等我確認。不要把 API 金鑰寫進命令或我會提交的檔案；問我金鑰應從哪裡取得。

檢視代理程式所依據的安裝指南

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

{
  "mcpServers": {
    "jaipilot": {
      "type": "http",
      "url": "https://api.jaipilot.com/functions/v1/jaipilot/mcp"
    }
  }
}

README

[JAIPilot]

JAIPilot — faster Java testing, stronger coverage, fewer bugs

Make your tests bulletproof.
~5× faster testing.
Enterprise grade unit tests. A 90%+ line coverage target.

Your Java testing specialist—straight from the terminal. Generate, repair, and verify tests. Fewer prompts. Stronger assertions.

Optimized for outcomes. Choose an outcome. Select your code. Get verified tests. Start with the CLI. Add MCP when you want JAIPilot inside your coding agent.

Download CLI · Get started · MCP setup · JAIPilot.com

See it run

Apache Kafka

[Actual Terminal recording of JAIPilot improving Apache Kafka's Deadline and LockUtils tests from 34.3% to 97.1% line coverage]

34.3% → 97.1% line coverage for two classes in 1m 53s. Added 13 tests; all 949 module tests passed.

Spring Petclinic — 100% coverage

[Actual Terminal recording of JAIPilot's coverage_100 workflow on Spring Petclinic, from 85.7% to 100% line coverage in 3m 21s]

85.7% → 100% line coverage across the project in 3m 21s. Added 36 tests with coverage_100; all 111 tests passed.

Production code preserved. Actual Terminal recordings; GIFs shorten idle waits. Reported times are full job durations with warm build caches; results vary. Commands, recordings, and full measurements →

How it works

Frontier LLM reasoning. Optimized testing tools. Prompts tailored to each outcome. Parallel workers in the managed service. Independent class tasks can run concurrently; local execution uses your toolchain and supported hardware parallelism.

  1. Inspect your code, build, and test framework.
  2. Generate or repair tests for behavior, boundaries, and failure paths, with meaningful assertions.
  3. Verify with focused tests, the final regression suite, and fresh coverage reports when configured.

90%+ line coverage target for generate_tests, improve_coverage, and lock_behavior, measured in selected production code when coverage is configured. Gaps and blockers are reported. Behavior baselines must pass before edits.

OutcomeWorkflow
Generate unit testsgenerate_tests
Improve coverageimprove_coverage
Fix failing testsfix_tests
Stabilize flaky testsstabilize_flaky_tests
Test current changestest_current_changes
Lock behavior before a refactorlock_behavior
Target 100% line coveragecoverage_100

Install

Requires Git, a JDK, and your project's build tools.

macOS / Linux

bash
curl -fsSL https://raw.githubusercontent.com/JAIPilot/jaipilot/main/install.sh | sh

Windows (PowerShell)

powershell
irm https://github.com/JAIPilot/jaipilot/releases/latest/download/install.ps1 | iex

npm

bash
npm install -g jaipilot

Homebrew (macOS / Linux)

bash
brew install JAIPilot/tap/jaipilot

Sign in, then run inside your Java repository:

bash
jaipilot auth loginjaipilot run improve_coverage --class com.acme.OrderService

Same JAIPilot account, subscription, and credits. Automatic update checks before workflows.

Install paths, manual updates, and troubleshooting →

Use the CLI

bash
jaipilot run generate_tests --class com.acme.OrderServicejaipilot run test_current_changes --alljaipilot run lock_behavior --path src/main/java/com/acme

Scope: whole repository, classes, paths, or selected lines. Review the result and git diff before accepting changes.

For a fresh, verified coverage target on a single-module Maven project:

bash
jaipilot run improve_coverage --all --coverage-target 80

The Maven adapter also provides local coverage checks and generation:

bash
./mvnw com.jaipilot:jaipilot-maven-plugin:1.2.1:check -Djaipilot.coverage.line=80./mvnw com.jaipilot:jaipilot-maven-plugin:1.2.1:run -Djaipilot.coverage.line=80

The Maven adapter 1.2.1 is published on Maven Central. Use mvnw.cmd in Windows PowerShell. Adapter setup, class/branch targets, and supported JaCoCo rules →

All commands, scopes, JSON output, and exit codes →

Why use JAIPilot if my coding agent already writes tests?

Fast testing. High coverage. Stronger tests. Less work for your coding agent.

  • ~5× faster testing than a general coding agent. Optimized prompts, tools, and parallel workers carry the testing workflow. Actual runtime varies by project. CLI examples: 97.1% coverage in 1m 53s on Kafka; 100% in 3m 21s on Petclinic.
  • 90%+ line coverage target. Meaningful assertions for selected behavior, boundaries, and failure paths, measured with your configured coverage tools.
  • Stronger regression protection. Lock behavior before edits; verify the original tests afterward. Our paired Codex example's saved tests caught 7/10 injected faults vs 3/10 with Codex alone.
  • Token efficiency. Offload test planning, generation, and repairs to JAIPilot's optimized service. This can reduce your coding agent's testing context, token use, and costs; JAIPilot credits apply.

Measured example: Codex with JAIPilot

Same refactor. Two fresh Codex sessions. Two Kafka utility classes. The MCP session used JAIPilot automatically, with no repository instruction file or extra testing prompt.

Saved regression suiteCodex aloneCodex + JAIPilot
Passing JUnit tests316
Line coverage of the selected Deadline class47.1%100%
Line coverage across both fixture classes40%65.7%
Branch coverage across both fixture classes30%80%
Injected faults caught by the saved tests3/107/10

13 more reusable tests. Four more deliberately introduced bugs caught. Same model, settings, initial tests, and warm build caches. One paired example; results vary.

The 90% target applies to selected code. Deadline: 100% (17/17 lines). The combined 65.7% (23/35) includes untargeted LockUtils, unchanged at 6/18 covered lines.

Method, evidence, and reproduction →

MCP for coding agents

Optional: make Codex, Claude Code, or another MCP agent better at Java testing. Give it a tested baseline before edits and real failures to guide repairs.

  1. Lock: lock_behavior creates and runs characterization tests before production edits, targeting 90%+ line coverage when configured.
  2. Edit: Your coding agent changes production code. Baseline tests stay unchanged.
  3. Verify: verify_behavior reruns the original baseline. Your agent investigates failures and repairs unintended changes before finishing.

Passing tests protect the behavior they exercise. Review old expectations for intentional behavior changes.

Codex

bash
codex mcp add jaipilot -- jaipilot mcp --repo /absolute/path/to/java-repository

Claude Code

bash
claude mcp add --transport stdio jaipilot -- jaipilot mcp --repo /absolute/path/to/java-repository

Start a new agent session. MCP instructions and tool descriptions include the workflow. No repository instruction file required. Tests and commands run locally. Jobs run in the background; 1 hour per local command by default.

Requires a build with the mcp command. Source setup, tool contracts, and timeout options →

Local execution and privacy

Builds and tests run locally. Selected source, project context, and command output go to JAIPilot's managed service and model provider. Sign-in credentials stay local with user-only permissions.

Privacy details · Ask a question · Report an issue · Report a vulnerability privately

MIT License · JAIPilot.com

來源：README.md，提交 d3fc8bc

工具

0
SourceWeft 尚未讀取此伺服器的工具。

版本歷史

1
  1. v7.2.1最新Sep 16, 2026

評分

已滿足適用檢查項的 85%

該評分衡量此收錄條目的完整程度，依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價，也不表示伺服器安全或目前可用：可連通性是帶日期的檢查結果，評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通必需25 分
  • 已宣告認證方式必需10 分
  • 已列出工具必需15 分

    請讓伺服器在無憑證時回應 tools/list，或發布可在沙箱中無憑證啟動的軟體包。

  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分

其他 Pilot 整合

QuickBooks Connector by Meridian

com.pilot.meridian官方發布者

讓助理透過託管的遠端連接器讀取與寫入 QuickBooks Online 公司資料。

Streamable HTTP可網頁執行驗證方式
商業與電商需要憑證Oct 24, 2026

AI Agent Connector by Meridian

com.pilot.meridian官方發布者

讓 AI 助理以授權的讀取或寫入權限連接 Xero，並可跨多個組織操作。

Streamable HTTP可網頁執行驗證方式
金融需要憑證Oct 24, 2026

XuLan ScreenPilot

由 GitHub 帳號 dahuam 發布

將真實的 Windows 螢幕轉換為帶編號的元素表，供你的 AI 代理程式操作。AI 翻譯

STDIO僅桌面
AI 與機器學習尚未驗證Nov 3, 2026

Covenant AAP - hosted developer pilot

由 GitHub 帳號 covenant-common 發布

託管的市場與合約起草伺服器：搜尋公開刊登、產生交易紀錄、起草與比對條款，並複核協議證據。

Streamable HTTP可網頁執行
商業與電商9 個工具Oct 31, 2026

Tag Pilot Search Console

由 GitHub 帳號 tagpilot 發布

唯讀的 Google Search Console MCP 伺服器，讓助理查詢你網站的點擊、曝光、CTR、排名、網站地圖與索引狀態。

STDIO僅桌面
商業與電商9 個工具Nov 3, 2026